1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * intel_pt_decoder.c: Intel Processor Trace support
4 * Copyright (c) 2013-2014, Intel Corporation.
5 */
6
7 #ifndef _GNU_SOURCE
8 #define _GNU_SOURCE
9 #endif
10 #include <stdlib.h>
11 #include <stdbool.h>
12 #include <string.h>
13 #include <errno.h>
14 #include <stdint.h>
15 #include <inttypes.h>
16 #include <linux/compiler.h>
17 #include <linux/string.h>
18 #include <linux/zalloc.h>
19
20 #include "../auxtrace.h"
21
22 #include "intel-pt-insn-decoder.h"
23 #include "intel-pt-pkt-decoder.h"
24 #include "intel-pt-decoder.h"
25 #include "intel-pt-log.h"
26
27 #define BITULL(x) (1ULL << (x))
28
29 /* IA32_RTIT_CTL MSR bits */
30 #define INTEL_PT_CYC_ENABLE BITULL(1)
31 #define INTEL_PT_CYC_THRESHOLD (BITULL(22) | BITULL(21) | BITULL(20) | BITULL(19))
32 #define INTEL_PT_CYC_THRESHOLD_SHIFT 19
33
34 #define INTEL_PT_BLK_SIZE 1024
35
36 #define BIT63 (((uint64_t)1 << 63))
37
38 #define SEVEN_BYTES 0xffffffffffffffULL
39
40 #define NO_VMCS 0xffffffffffULL
41
42 #define INTEL_PT_RETURN 1
43
44 /*
45 * Default maximum number of loops with no packets consumed i.e. stuck in a
46 * loop.
47 */
48 #define INTEL_PT_MAX_LOOPS 100000
49
50 struct intel_pt_blk {
51 struct intel_pt_blk *prev;
52 uint64_t ip[INTEL_PT_BLK_SIZE];
53 };
54
55 struct intel_pt_stack {
56 struct intel_pt_blk *blk;
57 struct intel_pt_blk *spare;
58 int pos;
59 };
60
61 enum intel_pt_p_once {
62 INTEL_PT_PRT_ONCE_UNK_VMCS,
63 INTEL_PT_PRT_ONCE_ERANGE,
64 };
65
66 enum intel_pt_pkt_state {
67 INTEL_PT_STATE_NO_PSB,
68 INTEL_PT_STATE_NO_IP,
69 INTEL_PT_STATE_ERR_RESYNC,
70 INTEL_PT_STATE_IN_SYNC,
71 INTEL_PT_STATE_TNT_CONT,
72 INTEL_PT_STATE_TNT,
73 INTEL_PT_STATE_TIP,
74 INTEL_PT_STATE_TIP_PGD,
75 INTEL_PT_STATE_FUP,
76 INTEL_PT_STATE_FUP_NO_TIP,
77 INTEL_PT_STATE_FUP_IN_PSB,
78 INTEL_PT_STATE_RESAMPLE,
79 INTEL_PT_STATE_VM_TIME_CORRELATION,
80 };
81
intel_pt_sample_time(enum intel_pt_pkt_state pkt_state)82 static inline bool intel_pt_sample_time(enum intel_pt_pkt_state pkt_state)
83 {
84 switch (pkt_state) {
85 case INTEL_PT_STATE_NO_PSB:
86 case INTEL_PT_STATE_NO_IP:
87 case INTEL_PT_STATE_ERR_RESYNC:
88 case INTEL_PT_STATE_IN_SYNC:
89 case INTEL_PT_STATE_TNT_CONT:
90 case INTEL_PT_STATE_RESAMPLE:
91 case INTEL_PT_STATE_VM_TIME_CORRELATION:
92 return true;
93 case INTEL_PT_STATE_TNT:
94 case INTEL_PT_STATE_TIP:
95 case INTEL_PT_STATE_TIP_PGD:
96 case INTEL_PT_STATE_FUP:
97 case INTEL_PT_STATE_FUP_NO_TIP:
98 case INTEL_PT_STATE_FUP_IN_PSB:
99 return false;
100 default:
101 return true;
102 };
103 }
104
105 #ifdef INTEL_PT_STRICT
106 #define INTEL_PT_STATE_ERR1 INTEL_PT_STATE_NO_PSB
107 #define INTEL_PT_STATE_ERR2 INTEL_PT_STATE_NO_PSB
108 #define INTEL_PT_STATE_ERR3 INTEL_PT_STATE_NO_PSB
109 #define INTEL_PT_STATE_ERR4 INTEL_PT_STATE_NO_PSB
110 #else
111 #define INTEL_PT_STATE_ERR1 (decoder->pkt_state)
112 #define INTEL_PT_STATE_ERR2 INTEL_PT_STATE_NO_IP
113 #define INTEL_PT_STATE_ERR3 INTEL_PT_STATE_ERR_RESYNC
114 #define INTEL_PT_STATE_ERR4 INTEL_PT_STATE_IN_SYNC
115 #endif
116
117 struct intel_pt_decoder {
118 int (*get_trace)(struct intel_pt_buffer *buffer, void *data);
119 int (*walk_insn)(struct intel_pt_insn *intel_pt_insn,
120 uint64_t *insn_cnt_ptr, uint64_t *ip, uint64_t to_ip,
121 uint64_t max_insn_cnt, void *data);
122 bool (*pgd_ip)(uint64_t ip, void *data);
123 int (*lookahead)(void *data, intel_pt_lookahead_cb_t cb, void *cb_data);
124 struct intel_pt_vmcs_info *(*findnew_vmcs_info)(void *data, uint64_t vmcs);
125 void *data;
126 struct intel_pt_state state;
127 const unsigned char *buf;
128 size_t len;
129 bool return_compression;
130 bool branch_enable;
131 bool mtc_insn;
132 bool pge;
133 bool have_tma;
134 bool have_cyc;
135 bool fixup_last_mtc;
136 bool have_last_ip;
137 bool in_psb;
138 bool hop;
139 bool leap;
140 bool vm_time_correlation;
141 bool vm_tm_corr_dry_run;
142 bool vm_tm_corr_reliable;
143 bool vm_tm_corr_same_buf;
144 bool vm_tm_corr_continuous;
145 bool nr;
146 bool next_nr;
147 enum intel_pt_param_flags flags;
148 uint64_t pos;
149 uint64_t last_ip;
150 uint64_t ip;
151 uint64_t pip_payload;
152 uint64_t timestamp;
153 uint64_t tsc_timestamp;
154 uint64_t ref_timestamp;
155 uint64_t buf_timestamp;
156 uint64_t sample_timestamp;
157 uint64_t ret_addr;
158 uint64_t ctc_timestamp;
159 uint64_t ctc_delta;
160 uint64_t cycle_cnt;
161 uint64_t cyc_ref_timestamp;
162 uint64_t first_timestamp;
163 uint64_t last_reliable_timestamp;
164 uint64_t vmcs;
165 uint64_t print_once;
166 uint64_t last_ctc;
167 uint32_t last_mtc;
168 uint32_t tsc_ctc_ratio_n;
169 uint32_t tsc_ctc_ratio_d;
170 uint32_t tsc_ctc_mult;
171 uint32_t tsc_slip;
172 uint32_t ctc_rem_mask;
173 int mtc_shift;
174 struct intel_pt_stack stack;
175 enum intel_pt_pkt_state pkt_state;
176 enum intel_pt_pkt_ctx pkt_ctx;
177 enum intel_pt_pkt_ctx prev_pkt_ctx;
178 enum intel_pt_blk_type blk_type;
179 int blk_type_pos;
180 struct intel_pt_pkt packet;
181 struct intel_pt_pkt tnt;
182 int pkt_step;
183 int pkt_len;
184 int last_packet_type;
185 unsigned int cbr;
186 unsigned int cbr_seen;
187 unsigned int max_non_turbo_ratio;
188 double max_non_turbo_ratio_fp;
189 double cbr_cyc_to_tsc;
190 double calc_cyc_to_tsc;
191 bool have_calc_cyc_to_tsc;
192 int exec_mode;
193 unsigned int insn_bytes;
194 uint64_t period;
195 enum intel_pt_period_type period_type;
196 uint64_t tot_insn_cnt;
197 uint64_t period_insn_cnt;
198 uint64_t period_mask;
199 uint64_t period_ticks;
200 uint64_t last_masked_timestamp;
201 uint64_t tot_cyc_cnt;
202 uint64_t sample_tot_cyc_cnt;
203 uint64_t base_cyc_cnt;
204 uint64_t cyc_cnt_timestamp;
205 uint64_t ctl;
206 uint64_t cyc_threshold;
207 double tsc_to_cyc;
208 bool continuous_period;
209 bool overflow;
210 bool set_fup_tx_flags;
211 bool set_fup_ptw;
212 bool set_fup_mwait;
213 bool set_fup_pwre;
214 bool set_fup_exstop;
215 bool set_fup_bep;
216 bool sample_cyc;
217 unsigned int fup_tx_flags;
218 unsigned int tx_flags;
219 uint64_t fup_ptw_payload;
220 uint64_t fup_mwait_payload;
221 uint64_t fup_pwre_payload;
222 uint64_t cbr_payload;
223 uint64_t timestamp_insn_cnt;
224 uint64_t sample_insn_cnt;
225 uint64_t stuck_ip;
226 int max_loops;
227 int no_progress;
228 int stuck_ip_prd;
229 int stuck_ip_cnt;
230 uint64_t psb_ip;
231 const unsigned char *next_buf;
232 size_t next_len;
233 unsigned char temp_buf[INTEL_PT_PKT_MAX_SZ];
234 };
235
intel_pt_lower_power_of_2(uint64_t x)236 static uint64_t intel_pt_lower_power_of_2(uint64_t x)
237 {
238 int i;
239
240 for (i = 0; x != 1; i++)
241 x >>= 1;
242
243 return x << i;
244 }
245
246 __printf(1, 2)
p_log(const char * fmt,...)247 static void p_log(const char *fmt, ...)
248 {
249 char buf[512];
250 va_list args;
251
252 va_start(args, fmt);
253 vsnprintf(buf, sizeof(buf), fmt, args);
254 va_end(args);
255
256 fprintf(stderr, "%s\n", buf);
257 intel_pt_log("%s\n", buf);
258 }
259
intel_pt_print_once(struct intel_pt_decoder * decoder,enum intel_pt_p_once id)260 static bool intel_pt_print_once(struct intel_pt_decoder *decoder,
261 enum intel_pt_p_once id)
262 {
263 uint64_t bit = 1ULL << id;
264
265 if (decoder->print_once & bit)
266 return false;
267 decoder->print_once |= bit;
268 return true;
269 }
270
intel_pt_cyc_threshold(uint64_t ctl)271 static uint64_t intel_pt_cyc_threshold(uint64_t ctl)
272 {
273 if (!(ctl & INTEL_PT_CYC_ENABLE))
274 return 0;
275
276 return (ctl & INTEL_PT_CYC_THRESHOLD) >> INTEL_PT_CYC_THRESHOLD_SHIFT;
277 }
278
intel_pt_setup_period(struct intel_pt_decoder * decoder)279 static void intel_pt_setup_period(struct intel_pt_decoder *decoder)
280 {
281 if (decoder->period_type == INTEL_PT_PERIOD_TICKS) {
282 uint64_t period;
283
284 period = intel_pt_lower_power_of_2(decoder->period);
285 decoder->period_mask = ~(period - 1);
286 decoder->period_ticks = period;
287 }
288 }
289
multdiv(uint64_t t,uint32_t n,uint32_t d)290 static uint64_t multdiv(uint64_t t, uint32_t n, uint32_t d)
291 {
292 if (!d)
293 return 0;
294 return (t / d) * n + ((t % d) * n) / d;
295 }
296
intel_pt_decoder_new(struct intel_pt_params * params)297 struct intel_pt_decoder *intel_pt_decoder_new(struct intel_pt_params *params)
298 {
299 struct intel_pt_decoder *decoder;
300
301 if (!params->get_trace || !params->walk_insn)
302 return NULL;
303
304 decoder = zalloc(sizeof(struct intel_pt_decoder));
305 if (!decoder)
306 return NULL;
307
308 decoder->get_trace = params->get_trace;
309 decoder->walk_insn = params->walk_insn;
310 decoder->pgd_ip = params->pgd_ip;
311 decoder->lookahead = params->lookahead;
312 decoder->findnew_vmcs_info = params->findnew_vmcs_info;
313 decoder->data = params->data;
314 decoder->return_compression = params->return_compression;
315 decoder->branch_enable = params->branch_enable;
316 decoder->hop = params->quick >= 1;
317 decoder->leap = params->quick >= 2;
318 decoder->vm_time_correlation = params->vm_time_correlation;
319 decoder->vm_tm_corr_dry_run = params->vm_tm_corr_dry_run;
320 decoder->first_timestamp = params->first_timestamp;
321 decoder->last_reliable_timestamp = params->first_timestamp;
322 decoder->max_loops = params->max_loops ? params->max_loops : INTEL_PT_MAX_LOOPS;
323
324 decoder->flags = params->flags;
325
326 decoder->ctl = params->ctl;
327 decoder->period = params->period;
328 decoder->period_type = params->period_type;
329
330 decoder->max_non_turbo_ratio = params->max_non_turbo_ratio;
331 decoder->max_non_turbo_ratio_fp = params->max_non_turbo_ratio;
332
333 decoder->cyc_threshold = intel_pt_cyc_threshold(decoder->ctl);
334
335 intel_pt_setup_period(decoder);
336
337 decoder->mtc_shift = params->mtc_period;
338 decoder->ctc_rem_mask = (1 << decoder->mtc_shift) - 1;
339
340 decoder->tsc_ctc_ratio_n = params->tsc_ctc_ratio_n;
341 decoder->tsc_ctc_ratio_d = params->tsc_ctc_ratio_d;
342
343 if (!decoder->tsc_ctc_ratio_n)
344 decoder->tsc_ctc_ratio_d = 0;
345
346 if (decoder->tsc_ctc_ratio_d) {
347 if (!(decoder->tsc_ctc_ratio_n % decoder->tsc_ctc_ratio_d))
348 decoder->tsc_ctc_mult = decoder->tsc_ctc_ratio_n /
349 decoder->tsc_ctc_ratio_d;
350 }
351
352 /*
353 * A TSC packet can slip past MTC packets so that the timestamp appears
354 * to go backwards. One estimate is that can be up to about 40 CPU
355 * cycles, which is certainly less than 0x1000 TSC ticks, but accept
356 * slippage an order of magnitude more to be on the safe side.
357 */
358 decoder->tsc_slip = 0x10000;
359
360 intel_pt_log("timestamp: mtc_shift %u\n", decoder->mtc_shift);
361 intel_pt_log("timestamp: tsc_ctc_ratio_n %u\n", decoder->tsc_ctc_ratio_n);
362 intel_pt_log("timestamp: tsc_ctc_ratio_d %u\n", decoder->tsc_ctc_ratio_d);
363 intel_pt_log("timestamp: tsc_ctc_mult %u\n", decoder->tsc_ctc_mult);
364 intel_pt_log("timestamp: tsc_slip %#x\n", decoder->tsc_slip);
365
366 if (decoder->hop)
367 intel_pt_log("Hop mode: decoding FUP and TIPs, but not TNT\n");
368
369 return decoder;
370 }
371
intel_pt_set_first_timestamp(struct intel_pt_decoder * decoder,uint64_t first_timestamp)372 void intel_pt_set_first_timestamp(struct intel_pt_decoder *decoder,
373 uint64_t first_timestamp)
374 {
375 decoder->first_timestamp = first_timestamp;
376 }
377
intel_pt_pop_blk(struct intel_pt_stack * stack)378 static void intel_pt_pop_blk(struct intel_pt_stack *stack)
379 {
380 struct intel_pt_blk *blk = stack->blk;
381
382 stack->blk = blk->prev;
383 if (!stack->spare)
384 stack->spare = blk;
385 else
386 free(blk);
387 }
388
intel_pt_pop(struct intel_pt_stack * stack)389 static uint64_t intel_pt_pop(struct intel_pt_stack *stack)
390 {
391 if (!stack->pos) {
392 if (!stack->blk)
393 return 0;
394 intel_pt_pop_blk(stack);
395 if (!stack->blk)
396 return 0;
397 stack->pos = INTEL_PT_BLK_SIZE;
398 }
399 return stack->blk->ip[--stack->pos];
400 }
401
intel_pt_alloc_blk(struct intel_pt_stack * stack)402 static int intel_pt_alloc_blk(struct intel_pt_stack *stack)
403 {
404 struct intel_pt_blk *blk;
405
406 if (stack->spare) {
407 blk = stack->spare;
408 stack->spare = NULL;
409 } else {
410 blk = malloc(sizeof(struct intel_pt_blk));
411 if (!blk)
412 return -ENOMEM;
413 }
414
415 blk->prev = stack->blk;
416 stack->blk = blk;
417 stack->pos = 0;
418 return 0;
419 }
420
intel_pt_push(struct intel_pt_stack * stack,uint64_t ip)421 static int intel_pt_push(struct intel_pt_stack *stack, uint64_t ip)
422 {
423 int err;
424
425 if (!stack->blk || stack->pos == INTEL_PT_BLK_SIZE) {
426 err = intel_pt_alloc_blk(stack);
427 if (err)
428 return err;
429 }
430
431 stack->blk->ip[stack->pos++] = ip;
432 return 0;
433 }
434
intel_pt_clear_stack(struct intel_pt_stack * stack)435 static void intel_pt_clear_stack(struct intel_pt_stack *stack)
436 {
437 while (stack->blk)
438 intel_pt_pop_blk(stack);
439 stack->pos = 0;
440 }
441
intel_pt_free_stack(struct intel_pt_stack * stack)442 static void intel_pt_free_stack(struct intel_pt_stack *stack)
443 {
444 intel_pt_clear_stack(stack);
445 zfree(&stack->blk);
446 zfree(&stack->spare);
447 }
448
intel_pt_decoder_free(struct intel_pt_decoder * decoder)449 void intel_pt_decoder_free(struct intel_pt_decoder *decoder)
450 {
451 intel_pt_free_stack(&decoder->stack);
452 free(decoder);
453 }
454
intel_pt_ext_err(int code)455 static int intel_pt_ext_err(int code)
456 {
457 switch (code) {
458 case -ENOMEM:
459 return INTEL_PT_ERR_NOMEM;
460 case -ENOSYS:
461 return INTEL_PT_ERR_INTERN;
462 case -EBADMSG:
463 return INTEL_PT_ERR_BADPKT;
464 case -ENODATA:
465 return INTEL_PT_ERR_NODATA;
466 case -EILSEQ:
467 return INTEL_PT_ERR_NOINSN;
468 case -ENOENT:
469 return INTEL_PT_ERR_MISMAT;
470 case -EOVERFLOW:
471 return INTEL_PT_ERR_OVR;
472 case -ENOSPC:
473 return INTEL_PT_ERR_LOST;
474 case -ELOOP:
475 return INTEL_PT_ERR_NELOOP;
476 default:
477 return INTEL_PT_ERR_UNK;
478 }
479 }
480
481 static const char *intel_pt_err_msgs[] = {
482 [INTEL_PT_ERR_NOMEM] = "Memory allocation failed",
483 [INTEL_PT_ERR_INTERN] = "Internal error",
484 [INTEL_PT_ERR_BADPKT] = "Bad packet",
485 [INTEL_PT_ERR_NODATA] = "No more data",
486 [INTEL_PT_ERR_NOINSN] = "Failed to get instruction",
487 [INTEL_PT_ERR_MISMAT] = "Trace doesn't match instruction",
488 [INTEL_PT_ERR_OVR] = "Overflow packet",
489 [INTEL_PT_ERR_LOST] = "Lost trace data",
490 [INTEL_PT_ERR_UNK] = "Unknown error!",
491 [INTEL_PT_ERR_NELOOP] = "Never-ending loop (refer perf config intel-pt.max-loops)",
492 };
493
intel_pt__strerror(int code,char * buf,size_t buflen)494 int intel_pt__strerror(int code, char *buf, size_t buflen)
495 {
496 if (code < 1 || code >= INTEL_PT_ERR_MAX)
497 code = INTEL_PT_ERR_UNK;
498 strlcpy(buf, intel_pt_err_msgs[code], buflen);
499 return 0;
500 }
501
intel_pt_calc_ip(const struct intel_pt_pkt * packet,uint64_t last_ip)502 static uint64_t intel_pt_calc_ip(const struct intel_pt_pkt *packet,
503 uint64_t last_ip)
504 {
505 uint64_t ip;
506
507 switch (packet->count) {
508 case 1:
509 ip = (last_ip & (uint64_t)0xffffffffffff0000ULL) |
510 packet->payload;
511 break;
512 case 2:
513 ip = (last_ip & (uint64_t)0xffffffff00000000ULL) |
514 packet->payload;
515 break;
516 case 3:
517 ip = packet->payload;
518 /* Sign-extend 6-byte ip */
519 if (ip & (uint64_t)0x800000000000ULL)
520 ip |= (uint64_t)0xffff000000000000ULL;
521 break;
522 case 4:
523 ip = (last_ip & (uint64_t)0xffff000000000000ULL) |
524 packet->payload;
525 break;
526 case 6:
527 ip = packet->payload;
528 break;
529 default:
530 return 0;
531 }
532
533 return ip;
534 }
535
intel_pt_set_last_ip(struct intel_pt_decoder * decoder)536 static inline void intel_pt_set_last_ip(struct intel_pt_decoder *decoder)
537 {
538 decoder->last_ip = intel_pt_calc_ip(&decoder->packet, decoder->last_ip);
539 decoder->have_last_ip = true;
540 }
541
intel_pt_set_ip(struct intel_pt_decoder * decoder)542 static inline void intel_pt_set_ip(struct intel_pt_decoder *decoder)
543 {
544 intel_pt_set_last_ip(decoder);
545 decoder->ip = decoder->last_ip;
546 }
547
intel_pt_decoder_log_packet(struct intel_pt_decoder * decoder)548 static void intel_pt_decoder_log_packet(struct intel_pt_decoder *decoder)
549 {
550 intel_pt_log_packet(&decoder->packet, decoder->pkt_len, decoder->pos,
551 decoder->buf);
552 }
553
intel_pt_bug(struct intel_pt_decoder * decoder)554 static int intel_pt_bug(struct intel_pt_decoder *decoder)
555 {
556 intel_pt_log("ERROR: Internal error\n");
557 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
558 return -ENOSYS;
559 }
560
intel_pt_clear_tx_flags(struct intel_pt_decoder * decoder)561 static inline void intel_pt_clear_tx_flags(struct intel_pt_decoder *decoder)
562 {
563 decoder->tx_flags = 0;
564 }
565
intel_pt_update_in_tx(struct intel_pt_decoder * decoder)566 static inline void intel_pt_update_in_tx(struct intel_pt_decoder *decoder)
567 {
568 decoder->tx_flags = decoder->packet.payload & INTEL_PT_IN_TX;
569 }
570
intel_pt_update_pip(struct intel_pt_decoder * decoder)571 static inline void intel_pt_update_pip(struct intel_pt_decoder *decoder)
572 {
573 decoder->pip_payload = decoder->packet.payload;
574 }
575
intel_pt_update_nr(struct intel_pt_decoder * decoder)576 static inline void intel_pt_update_nr(struct intel_pt_decoder *decoder)
577 {
578 decoder->next_nr = decoder->pip_payload & 1;
579 }
580
intel_pt_set_nr(struct intel_pt_decoder * decoder)581 static inline void intel_pt_set_nr(struct intel_pt_decoder *decoder)
582 {
583 decoder->nr = decoder->pip_payload & 1;
584 decoder->next_nr = decoder->nr;
585 }
586
intel_pt_set_pip(struct intel_pt_decoder * decoder)587 static inline void intel_pt_set_pip(struct intel_pt_decoder *decoder)
588 {
589 intel_pt_update_pip(decoder);
590 intel_pt_set_nr(decoder);
591 }
592
intel_pt_bad_packet(struct intel_pt_decoder * decoder)593 static int intel_pt_bad_packet(struct intel_pt_decoder *decoder)
594 {
595 intel_pt_clear_tx_flags(decoder);
596 decoder->have_tma = false;
597 decoder->pkt_len = 1;
598 decoder->pkt_step = 1;
599 intel_pt_decoder_log_packet(decoder);
600 if (decoder->pkt_state != INTEL_PT_STATE_NO_PSB) {
601 intel_pt_log("ERROR: Bad packet\n");
602 decoder->pkt_state = INTEL_PT_STATE_ERR1;
603 }
604 return -EBADMSG;
605 }
606
intel_pt_update_sample_time(struct intel_pt_decoder * decoder)607 static inline void intel_pt_update_sample_time(struct intel_pt_decoder *decoder)
608 {
609 decoder->sample_timestamp = decoder->timestamp;
610 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
611 decoder->state.cycles = decoder->tot_cyc_cnt;
612 }
613
intel_pt_reposition(struct intel_pt_decoder * decoder)614 static void intel_pt_reposition(struct intel_pt_decoder *decoder)
615 {
616 decoder->ip = 0;
617 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
618 decoder->timestamp = 0;
619 decoder->have_tma = false;
620 }
621
intel_pt_get_data(struct intel_pt_decoder * decoder,bool reposition)622 static int intel_pt_get_data(struct intel_pt_decoder *decoder, bool reposition)
623 {
624 struct intel_pt_buffer buffer = { .buf = 0, };
625 int ret;
626
627 decoder->pkt_step = 0;
628
629 intel_pt_log("Getting more data\n");
630 ret = decoder->get_trace(&buffer, decoder->data);
631 if (ret)
632 return ret;
633 decoder->buf = buffer.buf;
634 decoder->len = buffer.len;
635 if (!decoder->len) {
636 intel_pt_log("No more data\n");
637 return -ENODATA;
638 }
639 decoder->buf_timestamp = buffer.ref_timestamp;
640 if (!buffer.consecutive || reposition) {
641 intel_pt_reposition(decoder);
642 decoder->ref_timestamp = buffer.ref_timestamp;
643 decoder->state.trace_nr = buffer.trace_nr;
644 decoder->vm_tm_corr_same_buf = false;
645 intel_pt_log("Reference timestamp 0x%" PRIx64 "\n",
646 decoder->ref_timestamp);
647 return -ENOLINK;
648 }
649
650 return 0;
651 }
652
intel_pt_get_next_data(struct intel_pt_decoder * decoder,bool reposition)653 static int intel_pt_get_next_data(struct intel_pt_decoder *decoder,
654 bool reposition)
655 {
656 if (!decoder->next_buf)
657 return intel_pt_get_data(decoder, reposition);
658
659 decoder->buf = decoder->next_buf;
660 decoder->len = decoder->next_len;
661 decoder->next_buf = 0;
662 decoder->next_len = 0;
663 return 0;
664 }
665
intel_pt_get_split_packet(struct intel_pt_decoder * decoder)666 static int intel_pt_get_split_packet(struct intel_pt_decoder *decoder)
667 {
668 unsigned char *buf = decoder->temp_buf;
669 size_t old_len, len, n;
670 int ret;
671
672 old_len = decoder->len;
673 len = decoder->len;
674 memcpy(buf, decoder->buf, len);
675
676 ret = intel_pt_get_data(decoder, false);
677 if (ret) {
678 decoder->pos += old_len;
679 return ret < 0 ? ret : -EINVAL;
680 }
681
682 n = INTEL_PT_PKT_MAX_SZ - len;
683 if (n > decoder->len)
684 n = decoder->len;
685 memcpy(buf + len, decoder->buf, n);
686 len += n;
687
688 decoder->prev_pkt_ctx = decoder->pkt_ctx;
689 ret = intel_pt_get_packet(buf, len, &decoder->packet, &decoder->pkt_ctx);
690 if (ret < (int)old_len) {
691 decoder->next_buf = decoder->buf;
692 decoder->next_len = decoder->len;
693 decoder->buf = buf;
694 decoder->len = old_len;
695 return intel_pt_bad_packet(decoder);
696 }
697
698 decoder->next_buf = decoder->buf + (ret - old_len);
699 decoder->next_len = decoder->len - (ret - old_len);
700
701 decoder->buf = buf;
702 decoder->len = ret;
703
704 return ret;
705 }
706
707 struct intel_pt_pkt_info {
708 struct intel_pt_decoder *decoder;
709 struct intel_pt_pkt packet;
710 uint64_t pos;
711 int pkt_len;
712 int last_packet_type;
713 void *data;
714 };
715
716 typedef int (*intel_pt_pkt_cb_t)(struct intel_pt_pkt_info *pkt_info);
717
718 /* Lookahead packets in current buffer */
intel_pt_pkt_lookahead(struct intel_pt_decoder * decoder,intel_pt_pkt_cb_t cb,void * data)719 static int intel_pt_pkt_lookahead(struct intel_pt_decoder *decoder,
720 intel_pt_pkt_cb_t cb, void *data)
721 {
722 struct intel_pt_pkt_info pkt_info;
723 const unsigned char *buf = decoder->buf;
724 enum intel_pt_pkt_ctx pkt_ctx = decoder->pkt_ctx;
725 size_t len = decoder->len;
726 int ret;
727
728 pkt_info.decoder = decoder;
729 pkt_info.pos = decoder->pos;
730 pkt_info.pkt_len = decoder->pkt_step;
731 pkt_info.last_packet_type = decoder->last_packet_type;
732 pkt_info.data = data;
733
734 while (1) {
735 do {
736 pkt_info.pos += pkt_info.pkt_len;
737 buf += pkt_info.pkt_len;
738 len -= pkt_info.pkt_len;
739
740 if (!len)
741 return INTEL_PT_NEED_MORE_BYTES;
742
743 ret = intel_pt_get_packet(buf, len, &pkt_info.packet,
744 &pkt_ctx);
745 if (!ret)
746 return INTEL_PT_NEED_MORE_BYTES;
747 if (ret < 0)
748 return ret;
749
750 pkt_info.pkt_len = ret;
751 } while (pkt_info.packet.type == INTEL_PT_PAD);
752
753 ret = cb(&pkt_info);
754 if (ret)
755 return 0;
756
757 pkt_info.last_packet_type = pkt_info.packet.type;
758 }
759 }
760
761 struct intel_pt_calc_cyc_to_tsc_info {
762 uint64_t cycle_cnt;
763 unsigned int cbr;
764 uint32_t last_mtc;
765 uint64_t ctc_timestamp;
766 uint64_t ctc_delta;
767 uint64_t tsc_timestamp;
768 uint64_t timestamp;
769 bool have_tma;
770 bool fixup_last_mtc;
771 bool from_mtc;
772 double cbr_cyc_to_tsc;
773 };
774
775 /*
776 * MTC provides a 8-bit slice of CTC but the TMA packet only provides the lower
777 * 16 bits of CTC. If mtc_shift > 8 then some of the MTC bits are not in the CTC
778 * provided by the TMA packet. Fix-up the last_mtc calculated from the TMA
779 * packet by copying the missing bits from the current MTC assuming the least
780 * difference between the two, and that the current MTC comes after last_mtc.
781 */
intel_pt_fixup_last_mtc(uint32_t mtc,int mtc_shift,uint32_t * last_mtc)782 static void intel_pt_fixup_last_mtc(uint32_t mtc, int mtc_shift,
783 uint32_t *last_mtc)
784 {
785 uint32_t first_missing_bit = 1U << (16 - mtc_shift);
786 uint32_t mask = ~(first_missing_bit - 1);
787
788 *last_mtc |= mtc & mask;
789 if (*last_mtc >= mtc) {
790 *last_mtc -= first_missing_bit;
791 *last_mtc &= 0xff;
792 }
793 }
794
intel_pt_calc_cyc_cb(struct intel_pt_pkt_info * pkt_info)795 static int intel_pt_calc_cyc_cb(struct intel_pt_pkt_info *pkt_info)
796 {
797 struct intel_pt_decoder *decoder = pkt_info->decoder;
798 struct intel_pt_calc_cyc_to_tsc_info *data = pkt_info->data;
799 uint64_t timestamp;
800 double cyc_to_tsc;
801 unsigned int cbr;
802 uint32_t mtc, mtc_delta, ctc, fc, ctc_rem;
803
804 switch (pkt_info->packet.type) {
805 case INTEL_PT_TNT:
806 case INTEL_PT_TIP_PGE:
807 case INTEL_PT_TIP:
808 case INTEL_PT_FUP:
809 case INTEL_PT_PSB:
810 case INTEL_PT_PIP:
811 case INTEL_PT_MODE_EXEC:
812 case INTEL_PT_MODE_TSX:
813 case INTEL_PT_PSBEND:
814 case INTEL_PT_PAD:
815 case INTEL_PT_VMCS:
816 case INTEL_PT_MNT:
817 case INTEL_PT_PTWRITE:
818 case INTEL_PT_PTWRITE_IP:
819 case INTEL_PT_BBP:
820 case INTEL_PT_BIP:
821 case INTEL_PT_BEP:
822 case INTEL_PT_BEP_IP:
823 return 0;
824
825 case INTEL_PT_MTC:
826 if (!data->have_tma)
827 return 0;
828
829 mtc = pkt_info->packet.payload;
830 if (decoder->mtc_shift > 8 && data->fixup_last_mtc) {
831 data->fixup_last_mtc = false;
832 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
833 &data->last_mtc);
834 }
835 if (mtc > data->last_mtc)
836 mtc_delta = mtc - data->last_mtc;
837 else
838 mtc_delta = mtc + 256 - data->last_mtc;
839 data->ctc_delta += mtc_delta << decoder->mtc_shift;
840 data->last_mtc = mtc;
841
842 if (decoder->tsc_ctc_mult) {
843 timestamp = data->ctc_timestamp +
844 data->ctc_delta * decoder->tsc_ctc_mult;
845 } else {
846 timestamp = data->ctc_timestamp +
847 multdiv(data->ctc_delta,
848 decoder->tsc_ctc_ratio_n,
849 decoder->tsc_ctc_ratio_d);
850 }
851
852 if (timestamp < data->timestamp)
853 return 1;
854
855 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
856 data->timestamp = timestamp;
857 return 0;
858 }
859
860 break;
861
862 case INTEL_PT_TSC:
863 /*
864 * For now, do not support using TSC packets - refer
865 * intel_pt_calc_cyc_to_tsc().
866 */
867 if (data->from_mtc)
868 return 1;
869 timestamp = pkt_info->packet.payload |
870 (data->timestamp & (0xffULL << 56));
871 if (data->from_mtc && timestamp < data->timestamp &&
872 data->timestamp - timestamp < decoder->tsc_slip)
873 return 1;
874 if (timestamp < data->timestamp)
875 timestamp += (1ULL << 56);
876 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
877 if (data->from_mtc)
878 return 1;
879 data->tsc_timestamp = timestamp;
880 data->timestamp = timestamp;
881 return 0;
882 }
883 break;
884
885 case INTEL_PT_TMA:
886 if (data->from_mtc)
887 return 1;
888
889 if (!decoder->tsc_ctc_ratio_d)
890 return 0;
891
892 ctc = pkt_info->packet.payload;
893 fc = pkt_info->packet.count;
894 ctc_rem = ctc & decoder->ctc_rem_mask;
895
896 data->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
897
898 data->ctc_timestamp = data->tsc_timestamp - fc;
899 if (decoder->tsc_ctc_mult) {
900 data->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
901 } else {
902 data->ctc_timestamp -=
903 multdiv(ctc_rem, decoder->tsc_ctc_ratio_n,
904 decoder->tsc_ctc_ratio_d);
905 }
906
907 data->ctc_delta = 0;
908 data->have_tma = true;
909 data->fixup_last_mtc = true;
910
911 return 0;
912
913 case INTEL_PT_CYC:
914 data->cycle_cnt += pkt_info->packet.payload;
915 return 0;
916
917 case INTEL_PT_CBR:
918 cbr = pkt_info->packet.payload;
919 if (data->cbr && data->cbr != cbr)
920 return 1;
921 data->cbr = cbr;
922 data->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
923 return 0;
924
925 case INTEL_PT_TIP_PGD:
926 case INTEL_PT_TRACESTOP:
927 case INTEL_PT_EXSTOP:
928 case INTEL_PT_EXSTOP_IP:
929 case INTEL_PT_MWAIT:
930 case INTEL_PT_PWRE:
931 case INTEL_PT_PWRX:
932 case INTEL_PT_OVF:
933 case INTEL_PT_BAD: /* Does not happen */
934 default:
935 return 1;
936 }
937
938 if (!data->cbr && decoder->cbr) {
939 data->cbr = decoder->cbr;
940 data->cbr_cyc_to_tsc = decoder->cbr_cyc_to_tsc;
941 }
942
943 if (!data->cycle_cnt)
944 return 1;
945
946 cyc_to_tsc = (double)(timestamp - decoder->timestamp) / data->cycle_cnt;
947
948 if (data->cbr && cyc_to_tsc > data->cbr_cyc_to_tsc &&
949 cyc_to_tsc / data->cbr_cyc_to_tsc > 1.25) {
950 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle too big (c.f. CBR-based value %g), pos " x64_fmt "\n",
951 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
952 return 1;
953 }
954
955 decoder->calc_cyc_to_tsc = cyc_to_tsc;
956 decoder->have_calc_cyc_to_tsc = true;
957
958 if (data->cbr) {
959 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. CBR-based value %g, pos " x64_fmt "\n",
960 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
961 } else {
962 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. unknown CBR-based value, pos " x64_fmt "\n",
963 cyc_to_tsc, pkt_info->pos);
964 }
965
966 return 1;
967 }
968
intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder * decoder,bool from_mtc)969 static void intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder *decoder,
970 bool from_mtc)
971 {
972 struct intel_pt_calc_cyc_to_tsc_info data = {
973 .cycle_cnt = 0,
974 .cbr = 0,
975 .last_mtc = decoder->last_mtc,
976 .ctc_timestamp = decoder->ctc_timestamp,
977 .ctc_delta = decoder->ctc_delta,
978 .tsc_timestamp = decoder->tsc_timestamp,
979 .timestamp = decoder->timestamp,
980 .have_tma = decoder->have_tma,
981 .fixup_last_mtc = decoder->fixup_last_mtc,
982 .from_mtc = from_mtc,
983 .cbr_cyc_to_tsc = 0,
984 };
985
986 /*
987 * For now, do not support using TSC packets for at least the reasons:
988 * 1) timing might have stopped
989 * 2) TSC packets within PSB+ can slip against CYC packets
990 */
991 if (!from_mtc)
992 return;
993
994 intel_pt_pkt_lookahead(decoder, intel_pt_calc_cyc_cb, &data);
995 }
996
intel_pt_get_next_packet(struct intel_pt_decoder * decoder)997 static int intel_pt_get_next_packet(struct intel_pt_decoder *decoder)
998 {
999 int ret;
1000
1001 decoder->last_packet_type = decoder->packet.type;
1002
1003 do {
1004 decoder->pos += decoder->pkt_step;
1005 decoder->buf += decoder->pkt_step;
1006 decoder->len -= decoder->pkt_step;
1007
1008 if (!decoder->len) {
1009 ret = intel_pt_get_next_data(decoder, false);
1010 if (ret)
1011 return ret;
1012 }
1013
1014 decoder->prev_pkt_ctx = decoder->pkt_ctx;
1015 ret = intel_pt_get_packet(decoder->buf, decoder->len,
1016 &decoder->packet, &decoder->pkt_ctx);
1017 if (ret == INTEL_PT_NEED_MORE_BYTES && BITS_PER_LONG == 32 &&
1018 decoder->len < INTEL_PT_PKT_MAX_SZ && !decoder->next_buf) {
1019 ret = intel_pt_get_split_packet(decoder);
1020 if (ret < 0)
1021 return ret;
1022 }
1023 if (ret <= 0)
1024 return intel_pt_bad_packet(decoder);
1025
1026 decoder->pkt_len = ret;
1027 decoder->pkt_step = ret;
1028 intel_pt_decoder_log_packet(decoder);
1029 } while (decoder->packet.type == INTEL_PT_PAD);
1030
1031 return 0;
1032 }
1033
intel_pt_next_period(struct intel_pt_decoder * decoder)1034 static uint64_t intel_pt_next_period(struct intel_pt_decoder *decoder)
1035 {
1036 uint64_t timestamp, masked_timestamp;
1037
1038 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
1039 masked_timestamp = timestamp & decoder->period_mask;
1040 if (decoder->continuous_period) {
1041 if (masked_timestamp > decoder->last_masked_timestamp)
1042 return 1;
1043 } else {
1044 timestamp += 1;
1045 masked_timestamp = timestamp & decoder->period_mask;
1046 if (masked_timestamp > decoder->last_masked_timestamp) {
1047 decoder->last_masked_timestamp = masked_timestamp;
1048 decoder->continuous_period = true;
1049 }
1050 }
1051
1052 if (masked_timestamp < decoder->last_masked_timestamp)
1053 return decoder->period_ticks;
1054
1055 return decoder->period_ticks - (timestamp - masked_timestamp);
1056 }
1057
intel_pt_next_sample(struct intel_pt_decoder * decoder)1058 static uint64_t intel_pt_next_sample(struct intel_pt_decoder *decoder)
1059 {
1060 switch (decoder->period_type) {
1061 case INTEL_PT_PERIOD_INSTRUCTIONS:
1062 return decoder->period - decoder->period_insn_cnt;
1063 case INTEL_PT_PERIOD_TICKS:
1064 return intel_pt_next_period(decoder);
1065 case INTEL_PT_PERIOD_NONE:
1066 case INTEL_PT_PERIOD_MTC:
1067 default:
1068 return 0;
1069 }
1070 }
1071
intel_pt_sample_insn(struct intel_pt_decoder * decoder)1072 static void intel_pt_sample_insn(struct intel_pt_decoder *decoder)
1073 {
1074 uint64_t timestamp, masked_timestamp;
1075
1076 switch (decoder->period_type) {
1077 case INTEL_PT_PERIOD_INSTRUCTIONS:
1078 decoder->period_insn_cnt = 0;
1079 break;
1080 case INTEL_PT_PERIOD_TICKS:
1081 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
1082 masked_timestamp = timestamp & decoder->period_mask;
1083 if (masked_timestamp > decoder->last_masked_timestamp)
1084 decoder->last_masked_timestamp = masked_timestamp;
1085 else
1086 decoder->last_masked_timestamp += decoder->period_ticks;
1087 break;
1088 case INTEL_PT_PERIOD_NONE:
1089 case INTEL_PT_PERIOD_MTC:
1090 default:
1091 break;
1092 }
1093
1094 decoder->state.type |= INTEL_PT_INSTRUCTION;
1095 }
1096
intel_pt_walk_insn(struct intel_pt_decoder * decoder,struct intel_pt_insn * intel_pt_insn,uint64_t ip)1097 static int intel_pt_walk_insn(struct intel_pt_decoder *decoder,
1098 struct intel_pt_insn *intel_pt_insn, uint64_t ip)
1099 {
1100 uint64_t max_insn_cnt, insn_cnt = 0;
1101 int err;
1102
1103 if (!decoder->mtc_insn)
1104 decoder->mtc_insn = true;
1105
1106 max_insn_cnt = intel_pt_next_sample(decoder);
1107
1108 err = decoder->walk_insn(intel_pt_insn, &insn_cnt, &decoder->ip, ip,
1109 max_insn_cnt, decoder->data);
1110
1111 decoder->tot_insn_cnt += insn_cnt;
1112 decoder->timestamp_insn_cnt += insn_cnt;
1113 decoder->sample_insn_cnt += insn_cnt;
1114 decoder->period_insn_cnt += insn_cnt;
1115
1116 if (err) {
1117 decoder->no_progress = 0;
1118 decoder->pkt_state = INTEL_PT_STATE_ERR2;
1119 intel_pt_log_at("ERROR: Failed to get instruction",
1120 decoder->ip);
1121 if (err == -ENOENT)
1122 return -ENOLINK;
1123 return -EILSEQ;
1124 }
1125
1126 if (ip && decoder->ip == ip) {
1127 err = -EAGAIN;
1128 goto out;
1129 }
1130
1131 if (max_insn_cnt && insn_cnt >= max_insn_cnt)
1132 intel_pt_sample_insn(decoder);
1133
1134 if (intel_pt_insn->branch == INTEL_PT_BR_NO_BRANCH) {
1135 decoder->state.type = INTEL_PT_INSTRUCTION;
1136 decoder->state.from_ip = decoder->ip;
1137 decoder->state.to_ip = 0;
1138 decoder->ip += intel_pt_insn->length;
1139 err = INTEL_PT_RETURN;
1140 goto out;
1141 }
1142
1143 if (intel_pt_insn->op == INTEL_PT_OP_CALL) {
1144 /* Zero-length calls are excluded */
1145 if (intel_pt_insn->branch != INTEL_PT_BR_UNCONDITIONAL ||
1146 intel_pt_insn->rel) {
1147 err = intel_pt_push(&decoder->stack, decoder->ip +
1148 intel_pt_insn->length);
1149 if (err)
1150 goto out;
1151 }
1152 } else if (intel_pt_insn->op == INTEL_PT_OP_RET) {
1153 decoder->ret_addr = intel_pt_pop(&decoder->stack);
1154 }
1155
1156 if (intel_pt_insn->branch == INTEL_PT_BR_UNCONDITIONAL) {
1157 int cnt = decoder->no_progress++;
1158
1159 decoder->state.from_ip = decoder->ip;
1160 decoder->ip += intel_pt_insn->length +
1161 intel_pt_insn->rel;
1162 decoder->state.to_ip = decoder->ip;
1163 err = INTEL_PT_RETURN;
1164
1165 /*
1166 * Check for being stuck in a loop. This can happen if a
1167 * decoder error results in the decoder erroneously setting the
1168 * ip to an address that is itself in an infinite loop that
1169 * consumes no packets. When that happens, there must be an
1170 * unconditional branch.
1171 */
1172 if (cnt) {
1173 if (cnt == 1) {
1174 decoder->stuck_ip = decoder->state.to_ip;
1175 decoder->stuck_ip_prd = 1;
1176 decoder->stuck_ip_cnt = 1;
1177 } else if (cnt > decoder->max_loops ||
1178 decoder->state.to_ip == decoder->stuck_ip) {
1179 intel_pt_log_at("ERROR: Never-ending loop",
1180 decoder->state.to_ip);
1181 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1182 err = -ELOOP;
1183 goto out;
1184 } else if (!--decoder->stuck_ip_cnt) {
1185 decoder->stuck_ip_prd += 1;
1186 decoder->stuck_ip_cnt = decoder->stuck_ip_prd;
1187 decoder->stuck_ip = decoder->state.to_ip;
1188 }
1189 }
1190 goto out_no_progress;
1191 }
1192 out:
1193 decoder->no_progress = 0;
1194 out_no_progress:
1195 decoder->state.insn_op = intel_pt_insn->op;
1196 decoder->state.insn_len = intel_pt_insn->length;
1197 memcpy(decoder->state.insn, intel_pt_insn->buf,
1198 INTEL_PT_INSN_BUF_SZ);
1199
1200 if (decoder->tx_flags & INTEL_PT_IN_TX)
1201 decoder->state.flags |= INTEL_PT_IN_TX;
1202
1203 return err;
1204 }
1205
intel_pt_fup_event(struct intel_pt_decoder * decoder)1206 static bool intel_pt_fup_event(struct intel_pt_decoder *decoder)
1207 {
1208 enum intel_pt_sample_type type = decoder->state.type;
1209 bool ret = false;
1210
1211 decoder->state.type &= ~INTEL_PT_BRANCH;
1212
1213 if (decoder->set_fup_tx_flags) {
1214 decoder->set_fup_tx_flags = false;
1215 decoder->tx_flags = decoder->fup_tx_flags;
1216 decoder->state.type |= INTEL_PT_TRANSACTION;
1217 if (decoder->fup_tx_flags & INTEL_PT_ABORT_TX)
1218 decoder->state.type |= INTEL_PT_BRANCH;
1219 decoder->state.flags = decoder->fup_tx_flags;
1220 ret = true;
1221 }
1222 if (decoder->set_fup_ptw) {
1223 decoder->set_fup_ptw = false;
1224 decoder->state.type |= INTEL_PT_PTW;
1225 decoder->state.flags |= INTEL_PT_FUP_IP;
1226 decoder->state.ptw_payload = decoder->fup_ptw_payload;
1227 ret = true;
1228 }
1229 if (decoder->set_fup_mwait) {
1230 decoder->set_fup_mwait = false;
1231 decoder->state.type |= INTEL_PT_MWAIT_OP;
1232 decoder->state.mwait_payload = decoder->fup_mwait_payload;
1233 ret = true;
1234 }
1235 if (decoder->set_fup_pwre) {
1236 decoder->set_fup_pwre = false;
1237 decoder->state.type |= INTEL_PT_PWR_ENTRY;
1238 decoder->state.pwre_payload = decoder->fup_pwre_payload;
1239 ret = true;
1240 }
1241 if (decoder->set_fup_exstop) {
1242 decoder->set_fup_exstop = false;
1243 decoder->state.type |= INTEL_PT_EX_STOP;
1244 decoder->state.flags |= INTEL_PT_FUP_IP;
1245 ret = true;
1246 }
1247 if (decoder->set_fup_bep) {
1248 decoder->set_fup_bep = false;
1249 decoder->state.type |= INTEL_PT_BLK_ITEMS;
1250 ret = true;
1251 }
1252 if (decoder->overflow) {
1253 decoder->overflow = false;
1254 if (!ret && !decoder->pge) {
1255 if (decoder->hop) {
1256 decoder->state.type = 0;
1257 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
1258 }
1259 decoder->pge = true;
1260 decoder->state.type |= INTEL_PT_BRANCH | INTEL_PT_TRACE_BEGIN;
1261 decoder->state.from_ip = 0;
1262 decoder->state.to_ip = decoder->ip;
1263 return true;
1264 }
1265 }
1266 if (ret) {
1267 decoder->state.from_ip = decoder->ip;
1268 decoder->state.to_ip = 0;
1269 } else {
1270 decoder->state.type = type;
1271 }
1272 return ret;
1273 }
1274
intel_pt_fup_with_nlip(struct intel_pt_decoder * decoder,struct intel_pt_insn * intel_pt_insn,uint64_t ip,int err)1275 static inline bool intel_pt_fup_with_nlip(struct intel_pt_decoder *decoder,
1276 struct intel_pt_insn *intel_pt_insn,
1277 uint64_t ip, int err)
1278 {
1279 return decoder->flags & INTEL_PT_FUP_WITH_NLIP && !err &&
1280 intel_pt_insn->branch == INTEL_PT_BR_INDIRECT &&
1281 ip == decoder->ip + intel_pt_insn->length;
1282 }
1283
intel_pt_walk_fup(struct intel_pt_decoder * decoder)1284 static int intel_pt_walk_fup(struct intel_pt_decoder *decoder)
1285 {
1286 struct intel_pt_insn intel_pt_insn;
1287 uint64_t ip;
1288 int err;
1289
1290 ip = decoder->last_ip;
1291
1292 while (1) {
1293 err = intel_pt_walk_insn(decoder, &intel_pt_insn, ip);
1294 if (err == INTEL_PT_RETURN)
1295 return 0;
1296 if (err == -EAGAIN ||
1297 intel_pt_fup_with_nlip(decoder, &intel_pt_insn, ip, err)) {
1298 bool no_tip = decoder->pkt_state != INTEL_PT_STATE_FUP;
1299
1300 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1301 if (intel_pt_fup_event(decoder) && no_tip)
1302 return 0;
1303 return -EAGAIN;
1304 }
1305 decoder->set_fup_tx_flags = false;
1306 if (err)
1307 return err;
1308
1309 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1310 intel_pt_log_at("ERROR: Unexpected indirect branch",
1311 decoder->ip);
1312 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1313 return -ENOENT;
1314 }
1315
1316 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1317 intel_pt_log_at("ERROR: Unexpected conditional branch",
1318 decoder->ip);
1319 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1320 return -ENOENT;
1321 }
1322
1323 intel_pt_bug(decoder);
1324 }
1325 }
1326
intel_pt_walk_tip(struct intel_pt_decoder * decoder)1327 static int intel_pt_walk_tip(struct intel_pt_decoder *decoder)
1328 {
1329 struct intel_pt_insn intel_pt_insn;
1330 int err;
1331
1332 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1333 if (err == INTEL_PT_RETURN &&
1334 decoder->pgd_ip &&
1335 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1336 (decoder->state.type & INTEL_PT_BRANCH) &&
1337 decoder->pgd_ip(decoder->state.to_ip, decoder->data)) {
1338 /* Unconditional branch leaving filter region */
1339 decoder->no_progress = 0;
1340 decoder->pge = false;
1341 decoder->continuous_period = false;
1342 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1343 decoder->state.type |= INTEL_PT_TRACE_END;
1344 intel_pt_update_nr(decoder);
1345 return 0;
1346 }
1347 if (err == INTEL_PT_RETURN)
1348 return 0;
1349 if (err)
1350 return err;
1351
1352 intel_pt_update_nr(decoder);
1353
1354 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1355 if (decoder->pkt_state == INTEL_PT_STATE_TIP_PGD) {
1356 decoder->pge = false;
1357 decoder->continuous_period = false;
1358 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1359 decoder->state.from_ip = decoder->ip;
1360 if (decoder->packet.count == 0) {
1361 decoder->state.to_ip = 0;
1362 } else {
1363 decoder->state.to_ip = decoder->last_ip;
1364 decoder->ip = decoder->last_ip;
1365 }
1366 decoder->state.type |= INTEL_PT_TRACE_END;
1367 } else {
1368 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1369 decoder->state.from_ip = decoder->ip;
1370 if (decoder->packet.count == 0) {
1371 decoder->state.to_ip = 0;
1372 } else {
1373 decoder->state.to_ip = decoder->last_ip;
1374 decoder->ip = decoder->last_ip;
1375 }
1376 }
1377 return 0;
1378 }
1379
1380 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1381 uint64_t to_ip = decoder->ip + intel_pt_insn.length +
1382 intel_pt_insn.rel;
1383
1384 if (decoder->pgd_ip &&
1385 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1386 decoder->pgd_ip(to_ip, decoder->data)) {
1387 /* Conditional branch leaving filter region */
1388 decoder->pge = false;
1389 decoder->continuous_period = false;
1390 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1391 decoder->ip = to_ip;
1392 decoder->state.from_ip = decoder->ip;
1393 decoder->state.to_ip = to_ip;
1394 decoder->state.type |= INTEL_PT_TRACE_END;
1395 return 0;
1396 }
1397 intel_pt_log_at("ERROR: Conditional branch when expecting indirect branch",
1398 decoder->ip);
1399 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1400 return -ENOENT;
1401 }
1402
1403 return intel_pt_bug(decoder);
1404 }
1405
intel_pt_walk_tnt(struct intel_pt_decoder * decoder)1406 static int intel_pt_walk_tnt(struct intel_pt_decoder *decoder)
1407 {
1408 struct intel_pt_insn intel_pt_insn;
1409 int err;
1410
1411 while (1) {
1412 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1413 if (err == INTEL_PT_RETURN)
1414 return 0;
1415 if (err)
1416 return err;
1417
1418 if (intel_pt_insn.op == INTEL_PT_OP_RET) {
1419 if (!decoder->return_compression) {
1420 intel_pt_log_at("ERROR: RET when expecting conditional branch",
1421 decoder->ip);
1422 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1423 return -ENOENT;
1424 }
1425 if (!decoder->ret_addr) {
1426 intel_pt_log_at("ERROR: Bad RET compression (stack empty)",
1427 decoder->ip);
1428 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1429 return -ENOENT;
1430 }
1431 if (!(decoder->tnt.payload & BIT63)) {
1432 intel_pt_log_at("ERROR: Bad RET compression (TNT=N)",
1433 decoder->ip);
1434 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1435 return -ENOENT;
1436 }
1437 decoder->tnt.count -= 1;
1438 if (decoder->tnt.count)
1439 decoder->pkt_state = INTEL_PT_STATE_TNT_CONT;
1440 else
1441 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1442 decoder->tnt.payload <<= 1;
1443 decoder->state.from_ip = decoder->ip;
1444 decoder->ip = decoder->ret_addr;
1445 decoder->state.to_ip = decoder->ip;
1446 return 0;
1447 }
1448
1449 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1450 /* Handle deferred TIPs */
1451 err = intel_pt_get_next_packet(decoder);
1452 if (err)
1453 return err;
1454 if (decoder->packet.type != INTEL_PT_TIP ||
1455 decoder->packet.count == 0) {
1456 intel_pt_log_at("ERROR: Missing deferred TIP for indirect branch",
1457 decoder->ip);
1458 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1459 decoder->pkt_step = 0;
1460 return -ENOENT;
1461 }
1462 intel_pt_set_last_ip(decoder);
1463 decoder->state.from_ip = decoder->ip;
1464 decoder->state.to_ip = decoder->last_ip;
1465 decoder->ip = decoder->last_ip;
1466 intel_pt_update_nr(decoder);
1467 return 0;
1468 }
1469
1470 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1471 decoder->tnt.count -= 1;
1472 if (decoder->tnt.count)
1473 decoder->pkt_state = INTEL_PT_STATE_TNT_CONT;
1474 else
1475 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1476 if (decoder->tnt.payload & BIT63) {
1477 decoder->tnt.payload <<= 1;
1478 decoder->state.from_ip = decoder->ip;
1479 decoder->ip += intel_pt_insn.length +
1480 intel_pt_insn.rel;
1481 decoder->state.to_ip = decoder->ip;
1482 return 0;
1483 }
1484 /* Instruction sample for a non-taken branch */
1485 if (decoder->state.type & INTEL_PT_INSTRUCTION) {
1486 decoder->tnt.payload <<= 1;
1487 decoder->state.type = INTEL_PT_INSTRUCTION;
1488 decoder->state.from_ip = decoder->ip;
1489 decoder->state.to_ip = 0;
1490 decoder->ip += intel_pt_insn.length;
1491 return 0;
1492 }
1493 decoder->sample_cyc = false;
1494 decoder->ip += intel_pt_insn.length;
1495 if (!decoder->tnt.count) {
1496 intel_pt_update_sample_time(decoder);
1497 return -EAGAIN;
1498 }
1499 decoder->tnt.payload <<= 1;
1500 continue;
1501 }
1502
1503 return intel_pt_bug(decoder);
1504 }
1505 }
1506
intel_pt_mode_tsx(struct intel_pt_decoder * decoder,bool * no_tip)1507 static int intel_pt_mode_tsx(struct intel_pt_decoder *decoder, bool *no_tip)
1508 {
1509 unsigned int fup_tx_flags;
1510 int err;
1511
1512 fup_tx_flags = decoder->packet.payload &
1513 (INTEL_PT_IN_TX | INTEL_PT_ABORT_TX);
1514 err = intel_pt_get_next_packet(decoder);
1515 if (err)
1516 return err;
1517 if (decoder->packet.type == INTEL_PT_FUP) {
1518 decoder->fup_tx_flags = fup_tx_flags;
1519 decoder->set_fup_tx_flags = true;
1520 if (!(decoder->fup_tx_flags & INTEL_PT_ABORT_TX))
1521 *no_tip = true;
1522 } else {
1523 intel_pt_log_at("ERROR: Missing FUP after MODE.TSX",
1524 decoder->pos);
1525 intel_pt_update_in_tx(decoder);
1526 }
1527 return 0;
1528 }
1529
intel_pt_8b_tsc(uint64_t timestamp,uint64_t ref_timestamp)1530 static uint64_t intel_pt_8b_tsc(uint64_t timestamp, uint64_t ref_timestamp)
1531 {
1532 timestamp |= (ref_timestamp & (0xffULL << 56));
1533
1534 if (timestamp < ref_timestamp) {
1535 if (ref_timestamp - timestamp > (1ULL << 55))
1536 timestamp += (1ULL << 56);
1537 } else {
1538 if (timestamp - ref_timestamp > (1ULL << 55))
1539 timestamp -= (1ULL << 56);
1540 }
1541
1542 return timestamp;
1543 }
1544
1545 /* For use only when decoder->vm_time_correlation is true */
intel_pt_time_in_range(struct intel_pt_decoder * decoder,uint64_t timestamp)1546 static bool intel_pt_time_in_range(struct intel_pt_decoder *decoder,
1547 uint64_t timestamp)
1548 {
1549 uint64_t max_timestamp = decoder->buf_timestamp;
1550
1551 if (!max_timestamp) {
1552 max_timestamp = decoder->last_reliable_timestamp +
1553 0x400000000ULL;
1554 }
1555 return timestamp >= decoder->last_reliable_timestamp &&
1556 timestamp < decoder->buf_timestamp;
1557 }
1558
intel_pt_calc_tsc_timestamp(struct intel_pt_decoder * decoder)1559 static void intel_pt_calc_tsc_timestamp(struct intel_pt_decoder *decoder)
1560 {
1561 uint64_t timestamp;
1562 bool bad = false;
1563
1564 decoder->have_tma = false;
1565
1566 if (decoder->ref_timestamp) {
1567 timestamp = intel_pt_8b_tsc(decoder->packet.payload,
1568 decoder->ref_timestamp);
1569 decoder->tsc_timestamp = timestamp;
1570 decoder->timestamp = timestamp;
1571 decoder->ref_timestamp = 0;
1572 decoder->timestamp_insn_cnt = 0;
1573 } else if (decoder->timestamp) {
1574 timestamp = decoder->packet.payload |
1575 (decoder->timestamp & (0xffULL << 56));
1576 decoder->tsc_timestamp = timestamp;
1577 if (timestamp < decoder->timestamp &&
1578 decoder->timestamp - timestamp < decoder->tsc_slip) {
1579 intel_pt_log_to("Suppressing backwards timestamp",
1580 timestamp);
1581 timestamp = decoder->timestamp;
1582 }
1583 if (timestamp < decoder->timestamp) {
1584 if (!decoder->buf_timestamp ||
1585 (timestamp + (1ULL << 56) < decoder->buf_timestamp)) {
1586 intel_pt_log_to("Wraparound timestamp", timestamp);
1587 timestamp += (1ULL << 56);
1588 decoder->tsc_timestamp = timestamp;
1589 } else {
1590 intel_pt_log_to("Suppressing bad timestamp", timestamp);
1591 timestamp = decoder->timestamp;
1592 bad = true;
1593 }
1594 }
1595 if (decoder->vm_time_correlation &&
1596 (bad || !intel_pt_time_in_range(decoder, timestamp)) &&
1597 intel_pt_print_once(decoder, INTEL_PT_PRT_ONCE_ERANGE))
1598 p_log("Timestamp out of range");
1599 decoder->timestamp = timestamp;
1600 decoder->timestamp_insn_cnt = 0;
1601 }
1602
1603 if (decoder->last_packet_type == INTEL_PT_CYC) {
1604 decoder->cyc_ref_timestamp = decoder->timestamp;
1605 decoder->cycle_cnt = 0;
1606 decoder->have_calc_cyc_to_tsc = false;
1607 intel_pt_calc_cyc_to_tsc(decoder, false);
1608 }
1609
1610 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1611 }
1612
intel_pt_overflow(struct intel_pt_decoder * decoder)1613 static int intel_pt_overflow(struct intel_pt_decoder *decoder)
1614 {
1615 intel_pt_log("ERROR: Buffer overflow\n");
1616 intel_pt_clear_tx_flags(decoder);
1617 intel_pt_set_nr(decoder);
1618 decoder->timestamp_insn_cnt = 0;
1619 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1620 decoder->state.from_ip = decoder->ip;
1621 decoder->ip = 0;
1622 decoder->pge = false;
1623 decoder->set_fup_tx_flags = false;
1624 decoder->set_fup_ptw = false;
1625 decoder->set_fup_mwait = false;
1626 decoder->set_fup_pwre = false;
1627 decoder->set_fup_exstop = false;
1628 decoder->set_fup_bep = false;
1629 decoder->overflow = true;
1630 return -EOVERFLOW;
1631 }
1632
intel_pt_mtc_cyc_cnt_pge(struct intel_pt_decoder * decoder)1633 static inline void intel_pt_mtc_cyc_cnt_pge(struct intel_pt_decoder *decoder)
1634 {
1635 if (decoder->have_cyc)
1636 return;
1637
1638 decoder->cyc_cnt_timestamp = decoder->timestamp;
1639 decoder->base_cyc_cnt = decoder->tot_cyc_cnt;
1640 }
1641
intel_pt_mtc_cyc_cnt_cbr(struct intel_pt_decoder * decoder)1642 static inline void intel_pt_mtc_cyc_cnt_cbr(struct intel_pt_decoder *decoder)
1643 {
1644 decoder->tsc_to_cyc = decoder->cbr / decoder->max_non_turbo_ratio_fp;
1645
1646 if (decoder->pge)
1647 intel_pt_mtc_cyc_cnt_pge(decoder);
1648 }
1649
intel_pt_mtc_cyc_cnt_upd(struct intel_pt_decoder * decoder)1650 static inline void intel_pt_mtc_cyc_cnt_upd(struct intel_pt_decoder *decoder)
1651 {
1652 uint64_t tot_cyc_cnt, tsc_delta;
1653
1654 if (decoder->have_cyc)
1655 return;
1656
1657 decoder->sample_cyc = true;
1658
1659 if (!decoder->pge || decoder->timestamp <= decoder->cyc_cnt_timestamp)
1660 return;
1661
1662 tsc_delta = decoder->timestamp - decoder->cyc_cnt_timestamp;
1663 tot_cyc_cnt = tsc_delta * decoder->tsc_to_cyc + decoder->base_cyc_cnt;
1664
1665 if (tot_cyc_cnt > decoder->tot_cyc_cnt)
1666 decoder->tot_cyc_cnt = tot_cyc_cnt;
1667 }
1668
intel_pt_calc_tma(struct intel_pt_decoder * decoder)1669 static void intel_pt_calc_tma(struct intel_pt_decoder *decoder)
1670 {
1671 uint32_t ctc = decoder->packet.payload;
1672 uint32_t fc = decoder->packet.count;
1673 uint32_t ctc_rem = ctc & decoder->ctc_rem_mask;
1674
1675 if (!decoder->tsc_ctc_ratio_d)
1676 return;
1677
1678 if (decoder->pge && !decoder->in_psb)
1679 intel_pt_mtc_cyc_cnt_pge(decoder);
1680 else
1681 intel_pt_mtc_cyc_cnt_upd(decoder);
1682
1683 decoder->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
1684 decoder->last_ctc = ctc - ctc_rem;
1685 decoder->ctc_timestamp = decoder->tsc_timestamp - fc;
1686 if (decoder->tsc_ctc_mult) {
1687 decoder->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
1688 } else {
1689 decoder->ctc_timestamp -= multdiv(ctc_rem,
1690 decoder->tsc_ctc_ratio_n,
1691 decoder->tsc_ctc_ratio_d);
1692 }
1693 decoder->ctc_delta = 0;
1694 decoder->have_tma = true;
1695 decoder->fixup_last_mtc = true;
1696 intel_pt_log("CTC timestamp " x64_fmt " last MTC %#x CTC rem %#x\n",
1697 decoder->ctc_timestamp, decoder->last_mtc, ctc_rem);
1698 }
1699
intel_pt_calc_mtc_timestamp(struct intel_pt_decoder * decoder)1700 static void intel_pt_calc_mtc_timestamp(struct intel_pt_decoder *decoder)
1701 {
1702 uint64_t timestamp;
1703 uint32_t mtc, mtc_delta;
1704
1705 if (!decoder->have_tma)
1706 return;
1707
1708 mtc = decoder->packet.payload;
1709
1710 if (decoder->mtc_shift > 8 && decoder->fixup_last_mtc) {
1711 decoder->fixup_last_mtc = false;
1712 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
1713 &decoder->last_mtc);
1714 }
1715
1716 if (mtc > decoder->last_mtc)
1717 mtc_delta = mtc - decoder->last_mtc;
1718 else
1719 mtc_delta = mtc + 256 - decoder->last_mtc;
1720
1721 decoder->ctc_delta += mtc_delta << decoder->mtc_shift;
1722
1723 if (decoder->tsc_ctc_mult) {
1724 timestamp = decoder->ctc_timestamp +
1725 decoder->ctc_delta * decoder->tsc_ctc_mult;
1726 } else {
1727 timestamp = decoder->ctc_timestamp +
1728 multdiv(decoder->ctc_delta,
1729 decoder->tsc_ctc_ratio_n,
1730 decoder->tsc_ctc_ratio_d);
1731 }
1732
1733 if (timestamp < decoder->timestamp)
1734 intel_pt_log("Suppressing MTC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1735 timestamp, decoder->timestamp);
1736 else
1737 decoder->timestamp = timestamp;
1738
1739 intel_pt_mtc_cyc_cnt_upd(decoder);
1740
1741 decoder->timestamp_insn_cnt = 0;
1742 decoder->last_mtc = mtc;
1743
1744 if (decoder->last_packet_type == INTEL_PT_CYC) {
1745 decoder->cyc_ref_timestamp = decoder->timestamp;
1746 decoder->cycle_cnt = 0;
1747 decoder->have_calc_cyc_to_tsc = false;
1748 intel_pt_calc_cyc_to_tsc(decoder, true);
1749 }
1750
1751 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1752 }
1753
intel_pt_calc_cbr(struct intel_pt_decoder * decoder)1754 static void intel_pt_calc_cbr(struct intel_pt_decoder *decoder)
1755 {
1756 unsigned int cbr = decoder->packet.payload & 0xff;
1757
1758 decoder->cbr_payload = decoder->packet.payload;
1759
1760 if (decoder->cbr == cbr)
1761 return;
1762
1763 decoder->cbr = cbr;
1764 decoder->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
1765
1766 intel_pt_mtc_cyc_cnt_cbr(decoder);
1767 }
1768
intel_pt_calc_cyc_timestamp(struct intel_pt_decoder * decoder)1769 static void intel_pt_calc_cyc_timestamp(struct intel_pt_decoder *decoder)
1770 {
1771 uint64_t timestamp = decoder->cyc_ref_timestamp;
1772
1773 decoder->have_cyc = true;
1774
1775 decoder->cycle_cnt += decoder->packet.payload;
1776 if (decoder->pge)
1777 decoder->tot_cyc_cnt += decoder->packet.payload;
1778 decoder->sample_cyc = true;
1779
1780 if (!decoder->cyc_ref_timestamp)
1781 return;
1782
1783 if (decoder->have_calc_cyc_to_tsc)
1784 timestamp += decoder->cycle_cnt * decoder->calc_cyc_to_tsc;
1785 else if (decoder->cbr)
1786 timestamp += decoder->cycle_cnt * decoder->cbr_cyc_to_tsc;
1787 else
1788 return;
1789
1790 if (timestamp < decoder->timestamp)
1791 intel_pt_log("Suppressing CYC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1792 timestamp, decoder->timestamp);
1793 else
1794 decoder->timestamp = timestamp;
1795
1796 decoder->timestamp_insn_cnt = 0;
1797
1798 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1799 }
1800
intel_pt_bbp(struct intel_pt_decoder * decoder)1801 static void intel_pt_bbp(struct intel_pt_decoder *decoder)
1802 {
1803 if (decoder->prev_pkt_ctx == INTEL_PT_NO_CTX) {
1804 memset(decoder->state.items.mask, 0, sizeof(decoder->state.items.mask));
1805 decoder->state.items.is_32_bit = false;
1806 }
1807 decoder->blk_type = decoder->packet.payload;
1808 decoder->blk_type_pos = intel_pt_blk_type_pos(decoder->blk_type);
1809 if (decoder->blk_type == INTEL_PT_GP_REGS)
1810 decoder->state.items.is_32_bit = decoder->packet.count;
1811 if (decoder->blk_type_pos < 0) {
1812 intel_pt_log("WARNING: Unknown block type %u\n",
1813 decoder->blk_type);
1814 } else if (decoder->state.items.mask[decoder->blk_type_pos]) {
1815 intel_pt_log("WARNING: Duplicate block type %u\n",
1816 decoder->blk_type);
1817 }
1818 }
1819
intel_pt_bip(struct intel_pt_decoder * decoder)1820 static void intel_pt_bip(struct intel_pt_decoder *decoder)
1821 {
1822 uint32_t id = decoder->packet.count;
1823 uint32_t bit = 1 << id;
1824 int pos = decoder->blk_type_pos;
1825
1826 if (pos < 0 || id >= INTEL_PT_BLK_ITEM_ID_CNT) {
1827 intel_pt_log("WARNING: Unknown block item %u type %d\n",
1828 id, decoder->blk_type);
1829 return;
1830 }
1831
1832 if (decoder->state.items.mask[pos] & bit) {
1833 intel_pt_log("WARNING: Duplicate block item %u type %d\n",
1834 id, decoder->blk_type);
1835 }
1836
1837 decoder->state.items.mask[pos] |= bit;
1838 decoder->state.items.val[pos][id] = decoder->packet.payload;
1839 }
1840
1841 /* Walk PSB+ packets when already in sync. */
intel_pt_walk_psbend(struct intel_pt_decoder * decoder)1842 static int intel_pt_walk_psbend(struct intel_pt_decoder *decoder)
1843 {
1844 int err;
1845
1846 decoder->in_psb = true;
1847
1848 while (1) {
1849 err = intel_pt_get_next_packet(decoder);
1850 if (err)
1851 goto out;
1852
1853 switch (decoder->packet.type) {
1854 case INTEL_PT_PSBEND:
1855 err = 0;
1856 goto out;
1857
1858 case INTEL_PT_TIP_PGD:
1859 case INTEL_PT_TIP_PGE:
1860 case INTEL_PT_TIP:
1861 case INTEL_PT_TNT:
1862 case INTEL_PT_TRACESTOP:
1863 case INTEL_PT_BAD:
1864 case INTEL_PT_PSB:
1865 case INTEL_PT_PTWRITE:
1866 case INTEL_PT_PTWRITE_IP:
1867 case INTEL_PT_EXSTOP:
1868 case INTEL_PT_EXSTOP_IP:
1869 case INTEL_PT_MWAIT:
1870 case INTEL_PT_PWRE:
1871 case INTEL_PT_PWRX:
1872 case INTEL_PT_BBP:
1873 case INTEL_PT_BIP:
1874 case INTEL_PT_BEP:
1875 case INTEL_PT_BEP_IP:
1876 decoder->have_tma = false;
1877 intel_pt_log("ERROR: Unexpected packet\n");
1878 err = -EAGAIN;
1879 goto out;
1880
1881 case INTEL_PT_OVF:
1882 err = intel_pt_overflow(decoder);
1883 goto out;
1884
1885 case INTEL_PT_TSC:
1886 intel_pt_calc_tsc_timestamp(decoder);
1887 break;
1888
1889 case INTEL_PT_TMA:
1890 intel_pt_calc_tma(decoder);
1891 break;
1892
1893 case INTEL_PT_CBR:
1894 intel_pt_calc_cbr(decoder);
1895 break;
1896
1897 case INTEL_PT_MODE_EXEC:
1898 decoder->exec_mode = decoder->packet.payload;
1899 break;
1900
1901 case INTEL_PT_PIP:
1902 intel_pt_set_pip(decoder);
1903 break;
1904
1905 case INTEL_PT_FUP:
1906 decoder->pge = true;
1907 if (decoder->packet.count) {
1908 intel_pt_set_last_ip(decoder);
1909 decoder->psb_ip = decoder->last_ip;
1910 }
1911 break;
1912
1913 case INTEL_PT_MODE_TSX:
1914 intel_pt_update_in_tx(decoder);
1915 break;
1916
1917 case INTEL_PT_MTC:
1918 intel_pt_calc_mtc_timestamp(decoder);
1919 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1920 decoder->state.type |= INTEL_PT_INSTRUCTION;
1921 break;
1922
1923 case INTEL_PT_CYC:
1924 intel_pt_calc_cyc_timestamp(decoder);
1925 break;
1926
1927 case INTEL_PT_VMCS:
1928 case INTEL_PT_MNT:
1929 case INTEL_PT_PAD:
1930 default:
1931 break;
1932 }
1933 }
1934 out:
1935 decoder->in_psb = false;
1936
1937 return err;
1938 }
1939
intel_pt_walk_fup_tip(struct intel_pt_decoder * decoder)1940 static int intel_pt_walk_fup_tip(struct intel_pt_decoder *decoder)
1941 {
1942 int err;
1943
1944 if (decoder->tx_flags & INTEL_PT_ABORT_TX) {
1945 decoder->tx_flags = 0;
1946 decoder->state.flags &= ~INTEL_PT_IN_TX;
1947 decoder->state.flags |= INTEL_PT_ABORT_TX;
1948 } else {
1949 decoder->state.flags |= INTEL_PT_ASYNC;
1950 }
1951
1952 while (1) {
1953 err = intel_pt_get_next_packet(decoder);
1954 if (err)
1955 return err;
1956
1957 switch (decoder->packet.type) {
1958 case INTEL_PT_TNT:
1959 case INTEL_PT_FUP:
1960 case INTEL_PT_TRACESTOP:
1961 case INTEL_PT_PSB:
1962 case INTEL_PT_TSC:
1963 case INTEL_PT_TMA:
1964 case INTEL_PT_MODE_TSX:
1965 case INTEL_PT_BAD:
1966 case INTEL_PT_PSBEND:
1967 case INTEL_PT_PTWRITE:
1968 case INTEL_PT_PTWRITE_IP:
1969 case INTEL_PT_EXSTOP:
1970 case INTEL_PT_EXSTOP_IP:
1971 case INTEL_PT_MWAIT:
1972 case INTEL_PT_PWRE:
1973 case INTEL_PT_PWRX:
1974 case INTEL_PT_BBP:
1975 case INTEL_PT_BIP:
1976 case INTEL_PT_BEP:
1977 case INTEL_PT_BEP_IP:
1978 intel_pt_log("ERROR: Missing TIP after FUP\n");
1979 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1980 decoder->pkt_step = 0;
1981 return -ENOENT;
1982
1983 case INTEL_PT_CBR:
1984 intel_pt_calc_cbr(decoder);
1985 break;
1986
1987 case INTEL_PT_OVF:
1988 return intel_pt_overflow(decoder);
1989
1990 case INTEL_PT_TIP_PGD:
1991 decoder->state.from_ip = decoder->ip;
1992 if (decoder->packet.count == 0) {
1993 decoder->state.to_ip = 0;
1994 } else {
1995 intel_pt_set_ip(decoder);
1996 decoder->state.to_ip = decoder->ip;
1997 }
1998 decoder->pge = false;
1999 decoder->continuous_period = false;
2000 decoder->state.type |= INTEL_PT_TRACE_END;
2001 intel_pt_update_nr(decoder);
2002 return 0;
2003
2004 case INTEL_PT_TIP_PGE:
2005 decoder->pge = true;
2006 intel_pt_log("Omitting PGE ip " x64_fmt "\n",
2007 decoder->ip);
2008 decoder->state.from_ip = 0;
2009 if (decoder->packet.count == 0) {
2010 decoder->state.to_ip = 0;
2011 } else {
2012 intel_pt_set_ip(decoder);
2013 decoder->state.to_ip = decoder->ip;
2014 }
2015 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
2016 intel_pt_mtc_cyc_cnt_pge(decoder);
2017 intel_pt_set_nr(decoder);
2018 return 0;
2019
2020 case INTEL_PT_TIP:
2021 decoder->state.from_ip = decoder->ip;
2022 if (decoder->packet.count == 0) {
2023 decoder->state.to_ip = 0;
2024 } else {
2025 intel_pt_set_ip(decoder);
2026 decoder->state.to_ip = decoder->ip;
2027 }
2028 intel_pt_update_nr(decoder);
2029 return 0;
2030
2031 case INTEL_PT_PIP:
2032 intel_pt_update_pip(decoder);
2033 break;
2034
2035 case INTEL_PT_MTC:
2036 intel_pt_calc_mtc_timestamp(decoder);
2037 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
2038 decoder->state.type |= INTEL_PT_INSTRUCTION;
2039 break;
2040
2041 case INTEL_PT_CYC:
2042 intel_pt_calc_cyc_timestamp(decoder);
2043 break;
2044
2045 case INTEL_PT_MODE_EXEC:
2046 decoder->exec_mode = decoder->packet.payload;
2047 break;
2048
2049 case INTEL_PT_VMCS:
2050 case INTEL_PT_MNT:
2051 case INTEL_PT_PAD:
2052 break;
2053
2054 default:
2055 return intel_pt_bug(decoder);
2056 }
2057 }
2058 }
2059
intel_pt_resample(struct intel_pt_decoder * decoder)2060 static int intel_pt_resample(struct intel_pt_decoder *decoder)
2061 {
2062 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2063 decoder->state.type = INTEL_PT_INSTRUCTION;
2064 decoder->state.from_ip = decoder->ip;
2065 decoder->state.to_ip = 0;
2066 return 0;
2067 }
2068
2069 struct intel_pt_vm_tsc_info {
2070 struct intel_pt_pkt pip_packet;
2071 struct intel_pt_pkt vmcs_packet;
2072 struct intel_pt_pkt tma_packet;
2073 bool tsc, pip, vmcs, tma, psbend;
2074 uint64_t ctc_delta;
2075 uint64_t last_ctc;
2076 int max_lookahead;
2077 };
2078
2079 /* Lookahead and get the PIP, VMCS and TMA packets from PSB+ */
intel_pt_vm_psb_lookahead_cb(struct intel_pt_pkt_info * pkt_info)2080 static int intel_pt_vm_psb_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
2081 {
2082 struct intel_pt_vm_tsc_info *data = pkt_info->data;
2083
2084 switch (pkt_info->packet.type) {
2085 case INTEL_PT_PAD:
2086 case INTEL_PT_MNT:
2087 case INTEL_PT_MODE_EXEC:
2088 case INTEL_PT_MODE_TSX:
2089 case INTEL_PT_MTC:
2090 case INTEL_PT_FUP:
2091 case INTEL_PT_CYC:
2092 case INTEL_PT_CBR:
2093 break;
2094
2095 case INTEL_PT_TSC:
2096 data->tsc = true;
2097 break;
2098
2099 case INTEL_PT_TMA:
2100 data->tma_packet = pkt_info->packet;
2101 data->tma = true;
2102 break;
2103
2104 case INTEL_PT_PIP:
2105 data->pip_packet = pkt_info->packet;
2106 data->pip = true;
2107 break;
2108
2109 case INTEL_PT_VMCS:
2110 data->vmcs_packet = pkt_info->packet;
2111 data->vmcs = true;
2112 break;
2113
2114 case INTEL_PT_PSBEND:
2115 data->psbend = true;
2116 return 1;
2117
2118 case INTEL_PT_TIP_PGE:
2119 case INTEL_PT_PTWRITE:
2120 case INTEL_PT_PTWRITE_IP:
2121 case INTEL_PT_EXSTOP:
2122 case INTEL_PT_EXSTOP_IP:
2123 case INTEL_PT_MWAIT:
2124 case INTEL_PT_PWRE:
2125 case INTEL_PT_PWRX:
2126 case INTEL_PT_BBP:
2127 case INTEL_PT_BIP:
2128 case INTEL_PT_BEP:
2129 case INTEL_PT_BEP_IP:
2130 case INTEL_PT_OVF:
2131 case INTEL_PT_BAD:
2132 case INTEL_PT_TNT:
2133 case INTEL_PT_TIP_PGD:
2134 case INTEL_PT_TIP:
2135 case INTEL_PT_PSB:
2136 case INTEL_PT_TRACESTOP:
2137 default:
2138 return 1;
2139 }
2140
2141 return 0;
2142 }
2143
2144 struct intel_pt_ovf_fup_info {
2145 int max_lookahead;
2146 bool found;
2147 };
2148
2149 /* Lookahead to detect a FUP packet after OVF */
intel_pt_ovf_fup_lookahead_cb(struct intel_pt_pkt_info * pkt_info)2150 static int intel_pt_ovf_fup_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
2151 {
2152 struct intel_pt_ovf_fup_info *data = pkt_info->data;
2153
2154 if (pkt_info->packet.type == INTEL_PT_CYC ||
2155 pkt_info->packet.type == INTEL_PT_MTC ||
2156 pkt_info->packet.type == INTEL_PT_TSC)
2157 return !--(data->max_lookahead);
2158 data->found = pkt_info->packet.type == INTEL_PT_FUP;
2159 return 1;
2160 }
2161
intel_pt_ovf_fup_lookahead(struct intel_pt_decoder * decoder)2162 static bool intel_pt_ovf_fup_lookahead(struct intel_pt_decoder *decoder)
2163 {
2164 struct intel_pt_ovf_fup_info data = {
2165 .max_lookahead = 16,
2166 .found = false,
2167 };
2168
2169 intel_pt_pkt_lookahead(decoder, intel_pt_ovf_fup_lookahead_cb, &data);
2170 return data.found;
2171 }
2172
2173 /* Lookahead and get the TMA packet after TSC */
intel_pt_tma_lookahead_cb(struct intel_pt_pkt_info * pkt_info)2174 static int intel_pt_tma_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
2175 {
2176 struct intel_pt_vm_tsc_info *data = pkt_info->data;
2177
2178 if (pkt_info->packet.type == INTEL_PT_CYC ||
2179 pkt_info->packet.type == INTEL_PT_MTC)
2180 return !--(data->max_lookahead);
2181
2182 if (pkt_info->packet.type == INTEL_PT_TMA) {
2183 data->tma_packet = pkt_info->packet;
2184 data->tma = true;
2185 }
2186 return 1;
2187 }
2188
intel_pt_ctc_to_tsc(struct intel_pt_decoder * decoder,uint64_t ctc)2189 static uint64_t intel_pt_ctc_to_tsc(struct intel_pt_decoder *decoder, uint64_t ctc)
2190 {
2191 if (decoder->tsc_ctc_mult)
2192 return ctc * decoder->tsc_ctc_mult;
2193 else
2194 return multdiv(ctc, decoder->tsc_ctc_ratio_n, decoder->tsc_ctc_ratio_d);
2195 }
2196
intel_pt_calc_expected_tsc(struct intel_pt_decoder * decoder,uint32_t ctc,uint32_t fc,uint64_t last_ctc_timestamp,uint64_t ctc_delta,uint32_t last_ctc)2197 static uint64_t intel_pt_calc_expected_tsc(struct intel_pt_decoder *decoder,
2198 uint32_t ctc,
2199 uint32_t fc,
2200 uint64_t last_ctc_timestamp,
2201 uint64_t ctc_delta,
2202 uint32_t last_ctc)
2203 {
2204 /* Number of CTC ticks from last_ctc_timestamp to last_mtc */
2205 uint64_t last_mtc_ctc = last_ctc + ctc_delta;
2206 /*
2207 * Number of CTC ticks from there until current TMA packet. We would
2208 * expect last_mtc_ctc to be before ctc, but the TSC packet can slip
2209 * past an MTC, so a sign-extended value is used.
2210 */
2211 uint64_t delta = (int16_t)((uint16_t)ctc - (uint16_t)last_mtc_ctc);
2212 /* Total CTC ticks from last_ctc_timestamp to current TMA packet */
2213 uint64_t new_ctc_delta = ctc_delta + delta;
2214 uint64_t expected_tsc;
2215
2216 /*
2217 * Convert CTC ticks to TSC ticks, add the starting point
2218 * (last_ctc_timestamp) and the fast counter from the TMA packet.
2219 */
2220 expected_tsc = last_ctc_timestamp + intel_pt_ctc_to_tsc(decoder, new_ctc_delta) + fc;
2221
2222 if (intel_pt_enable_logging) {
2223 intel_pt_log_x64(last_mtc_ctc);
2224 intel_pt_log_x32(last_ctc);
2225 intel_pt_log_x64(ctc_delta);
2226 intel_pt_log_x64(delta);
2227 intel_pt_log_x32(ctc);
2228 intel_pt_log_x64(new_ctc_delta);
2229 intel_pt_log_x64(last_ctc_timestamp);
2230 intel_pt_log_x32(fc);
2231 intel_pt_log_x64(intel_pt_ctc_to_tsc(decoder, new_ctc_delta));
2232 intel_pt_log_x64(expected_tsc);
2233 }
2234
2235 return expected_tsc;
2236 }
2237
intel_pt_expected_tsc(struct intel_pt_decoder * decoder,struct intel_pt_vm_tsc_info * data)2238 static uint64_t intel_pt_expected_tsc(struct intel_pt_decoder *decoder,
2239 struct intel_pt_vm_tsc_info *data)
2240 {
2241 uint32_t ctc = data->tma_packet.payload;
2242 uint32_t fc = data->tma_packet.count;
2243
2244 return intel_pt_calc_expected_tsc(decoder, ctc, fc,
2245 decoder->ctc_timestamp,
2246 data->ctc_delta, data->last_ctc);
2247 }
2248
intel_pt_translate_vm_tsc(struct intel_pt_decoder * decoder,struct intel_pt_vmcs_info * vmcs_info)2249 static void intel_pt_translate_vm_tsc(struct intel_pt_decoder *decoder,
2250 struct intel_pt_vmcs_info *vmcs_info)
2251 {
2252 uint64_t payload = decoder->packet.payload;
2253
2254 /* VMX adds the TSC Offset, so subtract to get host TSC */
2255 decoder->packet.payload -= vmcs_info->tsc_offset;
2256 /* TSC packet has only 7 bytes */
2257 decoder->packet.payload &= SEVEN_BYTES;
2258
2259 /*
2260 * The buffer is mmapped from the data file, so this also updates the
2261 * data file.
2262 */
2263 if (!decoder->vm_tm_corr_dry_run)
2264 memcpy((void *)decoder->buf + 1, &decoder->packet.payload, 7);
2265
2266 intel_pt_log("Translated VM TSC %#" PRIx64 " -> %#" PRIx64
2267 " VMCS %#" PRIx64 " TSC Offset %#" PRIx64 "\n",
2268 payload, decoder->packet.payload, vmcs_info->vmcs,
2269 vmcs_info->tsc_offset);
2270 }
2271
intel_pt_translate_vm_tsc_offset(struct intel_pt_decoder * decoder,uint64_t tsc_offset)2272 static void intel_pt_translate_vm_tsc_offset(struct intel_pt_decoder *decoder,
2273 uint64_t tsc_offset)
2274 {
2275 struct intel_pt_vmcs_info vmcs_info = {
2276 .vmcs = NO_VMCS,
2277 .tsc_offset = tsc_offset
2278 };
2279
2280 intel_pt_translate_vm_tsc(decoder, &vmcs_info);
2281 }
2282
in_vm(uint64_t pip_payload)2283 static inline bool in_vm(uint64_t pip_payload)
2284 {
2285 return pip_payload & 1;
2286 }
2287
pip_in_vm(struct intel_pt_pkt * pip_packet)2288 static inline bool pip_in_vm(struct intel_pt_pkt *pip_packet)
2289 {
2290 return pip_packet->payload & 1;
2291 }
2292
intel_pt_print_vmcs_info(struct intel_pt_vmcs_info * vmcs_info)2293 static void intel_pt_print_vmcs_info(struct intel_pt_vmcs_info *vmcs_info)
2294 {
2295 p_log("VMCS: %#" PRIx64 " TSC Offset %#" PRIx64,
2296 vmcs_info->vmcs, vmcs_info->tsc_offset);
2297 }
2298
intel_pt_vm_tm_corr_psb(struct intel_pt_decoder * decoder,struct intel_pt_vm_tsc_info * data)2299 static void intel_pt_vm_tm_corr_psb(struct intel_pt_decoder *decoder,
2300 struct intel_pt_vm_tsc_info *data)
2301 {
2302 memset(data, 0, sizeof(*data));
2303 data->ctc_delta = decoder->ctc_delta;
2304 data->last_ctc = decoder->last_ctc;
2305 intel_pt_pkt_lookahead(decoder, intel_pt_vm_psb_lookahead_cb, data);
2306 if (data->tsc && !data->psbend)
2307 p_log("ERROR: PSB without PSBEND");
2308 decoder->in_psb = data->psbend;
2309 }
2310
intel_pt_vm_tm_corr_first_tsc(struct intel_pt_decoder * decoder,struct intel_pt_vm_tsc_info * data,struct intel_pt_vmcs_info * vmcs_info,uint64_t host_tsc)2311 static void intel_pt_vm_tm_corr_first_tsc(struct intel_pt_decoder *decoder,
2312 struct intel_pt_vm_tsc_info *data,
2313 struct intel_pt_vmcs_info *vmcs_info,
2314 uint64_t host_tsc)
2315 {
2316 if (!decoder->in_psb) {
2317 /* Can't happen */
2318 p_log("ERROR: First TSC is not in PSB+");
2319 }
2320
2321 if (data->pip) {
2322 if (pip_in_vm(&data->pip_packet)) { /* Guest */
2323 if (vmcs_info && vmcs_info->tsc_offset) {
2324 intel_pt_translate_vm_tsc(decoder, vmcs_info);
2325 decoder->vm_tm_corr_reliable = true;
2326 } else {
2327 p_log("ERROR: First TSC, unknown TSC Offset");
2328 }
2329 } else { /* Host */
2330 decoder->vm_tm_corr_reliable = true;
2331 }
2332 } else { /* Host or Guest */
2333 decoder->vm_tm_corr_reliable = false;
2334 if (intel_pt_time_in_range(decoder, host_tsc)) {
2335 /* Assume Host */
2336 } else {
2337 /* Assume Guest */
2338 if (vmcs_info && vmcs_info->tsc_offset)
2339 intel_pt_translate_vm_tsc(decoder, vmcs_info);
2340 else
2341 p_log("ERROR: First TSC, no PIP, unknown TSC Offset");
2342 }
2343 }
2344 }
2345
intel_pt_vm_tm_corr_tsc(struct intel_pt_decoder * decoder,struct intel_pt_vm_tsc_info * data)2346 static void intel_pt_vm_tm_corr_tsc(struct intel_pt_decoder *decoder,
2347 struct intel_pt_vm_tsc_info *data)
2348 {
2349 struct intel_pt_vmcs_info *vmcs_info;
2350 uint64_t tsc_offset = 0;
2351 uint64_t vmcs;
2352 bool reliable = true;
2353 uint64_t expected_tsc;
2354 uint64_t host_tsc;
2355 uint64_t ref_timestamp;
2356
2357 bool assign = false;
2358 bool assign_reliable = false;
2359
2360 /* Already have 'data' for the in_psb case */
2361 if (!decoder->in_psb) {
2362 memset(data, 0, sizeof(*data));
2363 data->ctc_delta = decoder->ctc_delta;
2364 data->last_ctc = decoder->last_ctc;
2365 data->max_lookahead = 16;
2366 intel_pt_pkt_lookahead(decoder, intel_pt_tma_lookahead_cb, data);
2367 if (decoder->pge) {
2368 data->pip = true;
2369 data->pip_packet.payload = decoder->pip_payload;
2370 }
2371 }
2372
2373 /* Calculations depend on having TMA packets */
2374 if (!data->tma) {
2375 p_log("ERROR: TSC without TMA");
2376 return;
2377 }
2378
2379 vmcs = data->vmcs ? data->vmcs_packet.payload : decoder->vmcs;
2380 if (vmcs == NO_VMCS)
2381 vmcs = 0;
2382
2383 vmcs_info = decoder->findnew_vmcs_info(decoder->data, vmcs);
2384
2385 ref_timestamp = decoder->timestamp ? decoder->timestamp : decoder->buf_timestamp;
2386 host_tsc = intel_pt_8b_tsc(decoder->packet.payload, ref_timestamp);
2387
2388 if (!decoder->ctc_timestamp) {
2389 intel_pt_vm_tm_corr_first_tsc(decoder, data, vmcs_info, host_tsc);
2390 return;
2391 }
2392
2393 expected_tsc = intel_pt_expected_tsc(decoder, data);
2394
2395 tsc_offset = host_tsc - expected_tsc;
2396
2397 /* Determine if TSC is from Host or Guest */
2398 if (data->pip) {
2399 if (pip_in_vm(&data->pip_packet)) { /* Guest */
2400 if (!vmcs_info) {
2401 /* PIP NR=1 without VMCS cannot happen */
2402 p_log("ERROR: Missing VMCS");
2403 intel_pt_translate_vm_tsc_offset(decoder, tsc_offset);
2404 decoder->vm_tm_corr_reliable = false;
2405 return;
2406 }
2407 } else { /* Host */
2408 decoder->last_reliable_timestamp = host_tsc;
2409 decoder->vm_tm_corr_reliable = true;
2410 return;
2411 }
2412 } else { /* Host or Guest */
2413 reliable = false; /* Host/Guest is a guess, so not reliable */
2414 if (decoder->in_psb) {
2415 if (!tsc_offset)
2416 return; /* Zero TSC Offset, assume Host */
2417 /*
2418 * TSC packet has only 7 bytes of TSC. We have no
2419 * information about the Guest's 8th byte, but it
2420 * doesn't matter because we only need 7 bytes.
2421 * Here, since the 8th byte is unreliable and
2422 * irrelevant, compare only 7 byes.
2423 */
2424 if (vmcs_info &&
2425 (tsc_offset & SEVEN_BYTES) ==
2426 (vmcs_info->tsc_offset & SEVEN_BYTES)) {
2427 /* Same TSC Offset as last VMCS, assume Guest */
2428 goto guest;
2429 }
2430 }
2431 /*
2432 * Check if the host_tsc is within the expected range.
2433 * Note, we could narrow the range more by looking ahead for
2434 * the next host TSC in the same buffer, but we don't bother to
2435 * do that because this is probably good enough.
2436 */
2437 if (host_tsc >= expected_tsc && intel_pt_time_in_range(decoder, host_tsc)) {
2438 /* Within expected range for Host TSC, assume Host */
2439 decoder->vm_tm_corr_reliable = false;
2440 return;
2441 }
2442 }
2443
2444 guest: /* Assuming Guest */
2445
2446 /* Determine whether to assign TSC Offset */
2447 if (vmcs_info && vmcs_info->vmcs) {
2448 if (vmcs_info->tsc_offset && vmcs_info->reliable) {
2449 assign = false;
2450 } else if (decoder->in_psb && data->pip && decoder->vm_tm_corr_reliable &&
2451 decoder->vm_tm_corr_continuous && decoder->vm_tm_corr_same_buf) {
2452 /* Continuous tracing, TSC in a PSB is not a time loss */
2453 assign = true;
2454 assign_reliable = true;
2455 } else if (decoder->in_psb && data->pip && decoder->vm_tm_corr_same_buf) {
2456 /*
2457 * Unlikely to be a time loss TSC in a PSB which is not
2458 * at the start of a buffer.
2459 */
2460 assign = true;
2461 assign_reliable = false;
2462 }
2463 }
2464
2465 /* Record VMCS TSC Offset */
2466 if (assign && (vmcs_info->tsc_offset != tsc_offset ||
2467 vmcs_info->reliable != assign_reliable)) {
2468 bool print = vmcs_info->tsc_offset != tsc_offset;
2469
2470 vmcs_info->tsc_offset = tsc_offset;
2471 vmcs_info->reliable = assign_reliable;
2472 if (print)
2473 intel_pt_print_vmcs_info(vmcs_info);
2474 }
2475
2476 /* Determine what TSC Offset to use */
2477 if (vmcs_info && vmcs_info->tsc_offset) {
2478 if (!vmcs_info->reliable)
2479 reliable = false;
2480 intel_pt_translate_vm_tsc(decoder, vmcs_info);
2481 } else {
2482 reliable = false;
2483 if (vmcs_info) {
2484 if (!vmcs_info->error_printed) {
2485 p_log("ERROR: Unknown TSC Offset for VMCS %#" PRIx64,
2486 vmcs_info->vmcs);
2487 vmcs_info->error_printed = true;
2488 }
2489 } else {
2490 if (intel_pt_print_once(decoder, INTEL_PT_PRT_ONCE_UNK_VMCS))
2491 p_log("ERROR: Unknown VMCS");
2492 }
2493 intel_pt_translate_vm_tsc_offset(decoder, tsc_offset);
2494 }
2495
2496 decoder->vm_tm_corr_reliable = reliable;
2497 }
2498
intel_pt_vm_tm_corr_pebs_tsc(struct intel_pt_decoder * decoder)2499 static void intel_pt_vm_tm_corr_pebs_tsc(struct intel_pt_decoder *decoder)
2500 {
2501 uint64_t host_tsc = decoder->packet.payload;
2502 uint64_t guest_tsc = decoder->packet.payload;
2503 struct intel_pt_vmcs_info *vmcs_info;
2504 uint64_t vmcs;
2505
2506 vmcs = decoder->vmcs;
2507 if (vmcs == NO_VMCS)
2508 vmcs = 0;
2509
2510 vmcs_info = decoder->findnew_vmcs_info(decoder->data, vmcs);
2511
2512 if (decoder->pge) {
2513 if (in_vm(decoder->pip_payload)) { /* Guest */
2514 if (!vmcs_info) {
2515 /* PIP NR=1 without VMCS cannot happen */
2516 p_log("ERROR: Missing VMCS");
2517 }
2518 } else { /* Host */
2519 return;
2520 }
2521 } else { /* Host or Guest */
2522 if (intel_pt_time_in_range(decoder, host_tsc)) {
2523 /* Within expected range for Host TSC, assume Host */
2524 return;
2525 }
2526 }
2527
2528 if (vmcs_info) {
2529 /* Translate Guest TSC to Host TSC */
2530 host_tsc = ((guest_tsc & SEVEN_BYTES) - vmcs_info->tsc_offset) & SEVEN_BYTES;
2531 host_tsc = intel_pt_8b_tsc(host_tsc, decoder->timestamp);
2532 intel_pt_log("Translated VM TSC %#" PRIx64 " -> %#" PRIx64
2533 " VMCS %#" PRIx64 " TSC Offset %#" PRIx64 "\n",
2534 guest_tsc, host_tsc, vmcs_info->vmcs,
2535 vmcs_info->tsc_offset);
2536 if (!intel_pt_time_in_range(decoder, host_tsc) &&
2537 intel_pt_print_once(decoder, INTEL_PT_PRT_ONCE_ERANGE))
2538 p_log("Timestamp out of range");
2539 } else {
2540 if (intel_pt_print_once(decoder, INTEL_PT_PRT_ONCE_UNK_VMCS))
2541 p_log("ERROR: Unknown VMCS");
2542 host_tsc = decoder->timestamp;
2543 }
2544
2545 decoder->packet.payload = host_tsc;
2546
2547 if (!decoder->vm_tm_corr_dry_run)
2548 memcpy((void *)decoder->buf + 1, &host_tsc, 8);
2549 }
2550
intel_pt_vm_time_correlation(struct intel_pt_decoder * decoder)2551 static int intel_pt_vm_time_correlation(struct intel_pt_decoder *decoder)
2552 {
2553 struct intel_pt_vm_tsc_info data = { .psbend = false };
2554 bool pge;
2555 int err;
2556
2557 if (decoder->in_psb)
2558 intel_pt_vm_tm_corr_psb(decoder, &data);
2559
2560 while (1) {
2561 err = intel_pt_get_next_packet(decoder);
2562 if (err == -ENOLINK)
2563 continue;
2564 if (err)
2565 break;
2566
2567 switch (decoder->packet.type) {
2568 case INTEL_PT_TIP_PGD:
2569 decoder->pge = false;
2570 decoder->vm_tm_corr_continuous = false;
2571 break;
2572
2573 case INTEL_PT_TNT:
2574 case INTEL_PT_TIP:
2575 case INTEL_PT_TIP_PGE:
2576 decoder->pge = true;
2577 break;
2578
2579 case INTEL_PT_OVF:
2580 decoder->in_psb = false;
2581 pge = decoder->pge;
2582 decoder->pge = intel_pt_ovf_fup_lookahead(decoder);
2583 if (pge != decoder->pge)
2584 intel_pt_log("Surprising PGE change in OVF!");
2585 if (!decoder->pge)
2586 decoder->vm_tm_corr_continuous = false;
2587 break;
2588
2589 case INTEL_PT_FUP:
2590 if (decoder->in_psb)
2591 decoder->pge = true;
2592 break;
2593
2594 case INTEL_PT_TRACESTOP:
2595 decoder->pge = false;
2596 decoder->vm_tm_corr_continuous = false;
2597 decoder->have_tma = false;
2598 break;
2599
2600 case INTEL_PT_PSB:
2601 intel_pt_vm_tm_corr_psb(decoder, &data);
2602 break;
2603
2604 case INTEL_PT_PIP:
2605 decoder->pip_payload = decoder->packet.payload;
2606 break;
2607
2608 case INTEL_PT_MTC:
2609 intel_pt_calc_mtc_timestamp(decoder);
2610 break;
2611
2612 case INTEL_PT_TSC:
2613 intel_pt_vm_tm_corr_tsc(decoder, &data);
2614 intel_pt_calc_tsc_timestamp(decoder);
2615 decoder->vm_tm_corr_same_buf = true;
2616 decoder->vm_tm_corr_continuous = decoder->pge;
2617 break;
2618
2619 case INTEL_PT_TMA:
2620 intel_pt_calc_tma(decoder);
2621 break;
2622
2623 case INTEL_PT_CYC:
2624 intel_pt_calc_cyc_timestamp(decoder);
2625 break;
2626
2627 case INTEL_PT_CBR:
2628 intel_pt_calc_cbr(decoder);
2629 break;
2630
2631 case INTEL_PT_PSBEND:
2632 decoder->in_psb = false;
2633 data.psbend = false;
2634 break;
2635
2636 case INTEL_PT_VMCS:
2637 if (decoder->packet.payload != NO_VMCS)
2638 decoder->vmcs = decoder->packet.payload;
2639 break;
2640
2641 case INTEL_PT_BBP:
2642 decoder->blk_type = decoder->packet.payload;
2643 break;
2644
2645 case INTEL_PT_BIP:
2646 if (decoder->blk_type == INTEL_PT_PEBS_BASIC &&
2647 decoder->packet.count == 2)
2648 intel_pt_vm_tm_corr_pebs_tsc(decoder);
2649 break;
2650
2651 case INTEL_PT_BEP:
2652 case INTEL_PT_BEP_IP:
2653 decoder->blk_type = 0;
2654 break;
2655
2656 case INTEL_PT_MODE_EXEC:
2657 case INTEL_PT_MODE_TSX:
2658 case INTEL_PT_MNT:
2659 case INTEL_PT_PAD:
2660 case INTEL_PT_PTWRITE_IP:
2661 case INTEL_PT_PTWRITE:
2662 case INTEL_PT_MWAIT:
2663 case INTEL_PT_PWRE:
2664 case INTEL_PT_EXSTOP_IP:
2665 case INTEL_PT_EXSTOP:
2666 case INTEL_PT_PWRX:
2667 case INTEL_PT_BAD: /* Does not happen */
2668 default:
2669 break;
2670 }
2671 }
2672
2673 return err;
2674 }
2675
2676 #define HOP_PROCESS 0
2677 #define HOP_IGNORE 1
2678 #define HOP_RETURN 2
2679 #define HOP_AGAIN 3
2680
2681 static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder);
2682
2683 /* Hop mode: Ignore TNT, do not walk code, but get ip from FUPs and TIPs */
intel_pt_hop_trace(struct intel_pt_decoder * decoder,bool * no_tip,int * err)2684 static int intel_pt_hop_trace(struct intel_pt_decoder *decoder, bool *no_tip, int *err)
2685 {
2686 *err = 0;
2687
2688 /* Leap from PSB to PSB, getting ip from FUP within PSB+ */
2689 if (decoder->leap && !decoder->in_psb && decoder->packet.type != INTEL_PT_PSB) {
2690 *err = intel_pt_scan_for_psb(decoder);
2691 if (*err)
2692 return HOP_RETURN;
2693 }
2694
2695 switch (decoder->packet.type) {
2696 case INTEL_PT_TNT:
2697 return HOP_IGNORE;
2698
2699 case INTEL_PT_TIP_PGD:
2700 decoder->pge = false;
2701 if (!decoder->packet.count) {
2702 intel_pt_set_nr(decoder);
2703 return HOP_IGNORE;
2704 }
2705 intel_pt_set_ip(decoder);
2706 decoder->state.type |= INTEL_PT_TRACE_END;
2707 decoder->state.from_ip = 0;
2708 decoder->state.to_ip = decoder->ip;
2709 intel_pt_update_nr(decoder);
2710 return HOP_RETURN;
2711
2712 case INTEL_PT_TIP:
2713 if (!decoder->packet.count) {
2714 intel_pt_set_nr(decoder);
2715 return HOP_IGNORE;
2716 }
2717 intel_pt_set_ip(decoder);
2718 decoder->state.type = INTEL_PT_INSTRUCTION;
2719 decoder->state.from_ip = decoder->ip;
2720 decoder->state.to_ip = 0;
2721 intel_pt_update_nr(decoder);
2722 return HOP_RETURN;
2723
2724 case INTEL_PT_FUP:
2725 if (!decoder->packet.count)
2726 return HOP_IGNORE;
2727 intel_pt_set_ip(decoder);
2728 if (decoder->set_fup_mwait || decoder->set_fup_pwre)
2729 *no_tip = true;
2730 if (!decoder->branch_enable || !decoder->pge)
2731 *no_tip = true;
2732 if (*no_tip) {
2733 decoder->state.type = INTEL_PT_INSTRUCTION;
2734 decoder->state.from_ip = decoder->ip;
2735 decoder->state.to_ip = 0;
2736 intel_pt_fup_event(decoder);
2737 return HOP_RETURN;
2738 }
2739 intel_pt_fup_event(decoder);
2740 decoder->state.type |= INTEL_PT_INSTRUCTION | INTEL_PT_BRANCH;
2741 *err = intel_pt_walk_fup_tip(decoder);
2742 if (!*err && decoder->state.to_ip)
2743 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
2744 return HOP_RETURN;
2745
2746 case INTEL_PT_PSB:
2747 decoder->state.psb_offset = decoder->pos;
2748 decoder->psb_ip = 0;
2749 decoder->last_ip = 0;
2750 decoder->have_last_ip = true;
2751 *err = intel_pt_walk_psbend(decoder);
2752 if (*err == -EAGAIN)
2753 return HOP_AGAIN;
2754 if (*err)
2755 return HOP_RETURN;
2756 decoder->state.type = INTEL_PT_PSB_EVT;
2757 if (decoder->psb_ip) {
2758 decoder->state.type |= INTEL_PT_INSTRUCTION;
2759 decoder->ip = decoder->psb_ip;
2760 }
2761 decoder->state.from_ip = decoder->psb_ip;
2762 decoder->state.to_ip = 0;
2763 return HOP_RETURN;
2764
2765 case INTEL_PT_BAD:
2766 case INTEL_PT_PAD:
2767 case INTEL_PT_TIP_PGE:
2768 case INTEL_PT_TSC:
2769 case INTEL_PT_TMA:
2770 case INTEL_PT_MODE_EXEC:
2771 case INTEL_PT_MODE_TSX:
2772 case INTEL_PT_MTC:
2773 case INTEL_PT_CYC:
2774 case INTEL_PT_VMCS:
2775 case INTEL_PT_PSBEND:
2776 case INTEL_PT_CBR:
2777 case INTEL_PT_TRACESTOP:
2778 case INTEL_PT_PIP:
2779 case INTEL_PT_OVF:
2780 case INTEL_PT_MNT:
2781 case INTEL_PT_PTWRITE:
2782 case INTEL_PT_PTWRITE_IP:
2783 case INTEL_PT_EXSTOP:
2784 case INTEL_PT_EXSTOP_IP:
2785 case INTEL_PT_MWAIT:
2786 case INTEL_PT_PWRE:
2787 case INTEL_PT_PWRX:
2788 case INTEL_PT_BBP:
2789 case INTEL_PT_BIP:
2790 case INTEL_PT_BEP:
2791 case INTEL_PT_BEP_IP:
2792 default:
2793 return HOP_PROCESS;
2794 }
2795 }
2796
2797 struct intel_pt_psb_info {
2798 struct intel_pt_pkt fup_packet;
2799 bool fup;
2800 int after_psbend;
2801 };
2802
2803 /* Lookahead and get the FUP packet from PSB+ */
intel_pt_psb_lookahead_cb(struct intel_pt_pkt_info * pkt_info)2804 static int intel_pt_psb_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
2805 {
2806 struct intel_pt_psb_info *data = pkt_info->data;
2807
2808 switch (pkt_info->packet.type) {
2809 case INTEL_PT_PAD:
2810 case INTEL_PT_MNT:
2811 case INTEL_PT_TSC:
2812 case INTEL_PT_TMA:
2813 case INTEL_PT_MODE_EXEC:
2814 case INTEL_PT_MODE_TSX:
2815 case INTEL_PT_MTC:
2816 case INTEL_PT_CYC:
2817 case INTEL_PT_VMCS:
2818 case INTEL_PT_CBR:
2819 case INTEL_PT_PIP:
2820 if (data->after_psbend) {
2821 data->after_psbend -= 1;
2822 if (!data->after_psbend)
2823 return 1;
2824 }
2825 break;
2826
2827 case INTEL_PT_FUP:
2828 if (data->after_psbend)
2829 return 1;
2830 if (data->fup || pkt_info->packet.count == 0)
2831 return 1;
2832 data->fup_packet = pkt_info->packet;
2833 data->fup = true;
2834 break;
2835
2836 case INTEL_PT_PSBEND:
2837 if (!data->fup)
2838 return 1;
2839 /* Keep going to check for a TIP.PGE */
2840 data->after_psbend = 6;
2841 break;
2842
2843 case INTEL_PT_TIP_PGE:
2844 /* Ignore FUP in PSB+ if followed by TIP.PGE */
2845 if (data->after_psbend)
2846 data->fup = false;
2847 return 1;
2848
2849 case INTEL_PT_PTWRITE:
2850 case INTEL_PT_PTWRITE_IP:
2851 case INTEL_PT_EXSTOP:
2852 case INTEL_PT_EXSTOP_IP:
2853 case INTEL_PT_MWAIT:
2854 case INTEL_PT_PWRE:
2855 case INTEL_PT_PWRX:
2856 case INTEL_PT_BBP:
2857 case INTEL_PT_BIP:
2858 case INTEL_PT_BEP:
2859 case INTEL_PT_BEP_IP:
2860 if (data->after_psbend) {
2861 data->after_psbend -= 1;
2862 if (!data->after_psbend)
2863 return 1;
2864 break;
2865 }
2866 return 1;
2867
2868 case INTEL_PT_OVF:
2869 case INTEL_PT_BAD:
2870 case INTEL_PT_TNT:
2871 case INTEL_PT_TIP_PGD:
2872 case INTEL_PT_TIP:
2873 case INTEL_PT_PSB:
2874 case INTEL_PT_TRACESTOP:
2875 default:
2876 return 1;
2877 }
2878
2879 return 0;
2880 }
2881
intel_pt_psb(struct intel_pt_decoder * decoder)2882 static int intel_pt_psb(struct intel_pt_decoder *decoder)
2883 {
2884 int err;
2885
2886 decoder->last_ip = 0;
2887 decoder->psb_ip = 0;
2888 decoder->have_last_ip = true;
2889 intel_pt_clear_stack(&decoder->stack);
2890 err = intel_pt_walk_psbend(decoder);
2891 if (err)
2892 return err;
2893 decoder->state.type = INTEL_PT_PSB_EVT;
2894 decoder->state.from_ip = decoder->psb_ip;
2895 decoder->state.to_ip = 0;
2896 return 0;
2897 }
2898
intel_pt_fup_in_psb(struct intel_pt_decoder * decoder)2899 static int intel_pt_fup_in_psb(struct intel_pt_decoder *decoder)
2900 {
2901 int err;
2902
2903 if (decoder->ip != decoder->last_ip) {
2904 err = intel_pt_walk_fup(decoder);
2905 if (!err || err != -EAGAIN)
2906 return err;
2907 }
2908
2909 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2910 err = intel_pt_psb(decoder);
2911 if (err) {
2912 decoder->pkt_state = INTEL_PT_STATE_ERR3;
2913 return -ENOENT;
2914 }
2915
2916 return 0;
2917 }
2918
intel_pt_psb_with_fup(struct intel_pt_decoder * decoder,int * err)2919 static bool intel_pt_psb_with_fup(struct intel_pt_decoder *decoder, int *err)
2920 {
2921 struct intel_pt_psb_info data = { .fup = false };
2922
2923 if (!decoder->branch_enable)
2924 return false;
2925
2926 intel_pt_pkt_lookahead(decoder, intel_pt_psb_lookahead_cb, &data);
2927 if (!data.fup)
2928 return false;
2929
2930 decoder->packet = data.fup_packet;
2931 intel_pt_set_last_ip(decoder);
2932 decoder->pkt_state = INTEL_PT_STATE_FUP_IN_PSB;
2933
2934 *err = intel_pt_fup_in_psb(decoder);
2935
2936 return true;
2937 }
2938
intel_pt_walk_trace(struct intel_pt_decoder * decoder)2939 static int intel_pt_walk_trace(struct intel_pt_decoder *decoder)
2940 {
2941 int last_packet_type = INTEL_PT_PAD;
2942 bool no_tip = false;
2943 int err;
2944
2945 while (1) {
2946 err = intel_pt_get_next_packet(decoder);
2947 if (err)
2948 return err;
2949 next:
2950 err = 0;
2951 if (decoder->cyc_threshold) {
2952 if (decoder->sample_cyc && last_packet_type != INTEL_PT_CYC)
2953 decoder->sample_cyc = false;
2954 last_packet_type = decoder->packet.type;
2955 }
2956
2957 if (decoder->hop) {
2958 switch (intel_pt_hop_trace(decoder, &no_tip, &err)) {
2959 case HOP_IGNORE:
2960 continue;
2961 case HOP_RETURN:
2962 return err;
2963 case HOP_AGAIN:
2964 goto next;
2965 default:
2966 break;
2967 }
2968 }
2969
2970 switch (decoder->packet.type) {
2971 case INTEL_PT_TNT:
2972 if (!decoder->packet.count)
2973 break;
2974 decoder->tnt = decoder->packet;
2975 decoder->pkt_state = INTEL_PT_STATE_TNT;
2976 err = intel_pt_walk_tnt(decoder);
2977 if (err == -EAGAIN)
2978 break;
2979 return err;
2980
2981 case INTEL_PT_TIP_PGD:
2982 if (decoder->packet.count != 0)
2983 intel_pt_set_last_ip(decoder);
2984 decoder->pkt_state = INTEL_PT_STATE_TIP_PGD;
2985 return intel_pt_walk_tip(decoder);
2986
2987 case INTEL_PT_TIP_PGE: {
2988 decoder->pge = true;
2989 decoder->overflow = false;
2990 intel_pt_mtc_cyc_cnt_pge(decoder);
2991 intel_pt_set_nr(decoder);
2992 if (decoder->packet.count == 0) {
2993 intel_pt_log_at("Skipping zero TIP.PGE",
2994 decoder->pos);
2995 break;
2996 }
2997 intel_pt_set_ip(decoder);
2998 decoder->state.from_ip = 0;
2999 decoder->state.to_ip = decoder->ip;
3000 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
3001 /*
3002 * In hop mode, resample to get the to_ip as an
3003 * "instruction" sample.
3004 */
3005 if (decoder->hop)
3006 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
3007 return 0;
3008 }
3009
3010 case INTEL_PT_OVF:
3011 return intel_pt_overflow(decoder);
3012
3013 case INTEL_PT_TIP:
3014 if (decoder->packet.count != 0)
3015 intel_pt_set_last_ip(decoder);
3016 decoder->pkt_state = INTEL_PT_STATE_TIP;
3017 return intel_pt_walk_tip(decoder);
3018
3019 case INTEL_PT_FUP:
3020 if (decoder->packet.count == 0) {
3021 intel_pt_log_at("Skipping zero FUP",
3022 decoder->pos);
3023 no_tip = false;
3024 break;
3025 }
3026 intel_pt_set_last_ip(decoder);
3027 if (!decoder->branch_enable || !decoder->pge) {
3028 decoder->ip = decoder->last_ip;
3029 if (intel_pt_fup_event(decoder))
3030 return 0;
3031 no_tip = false;
3032 break;
3033 }
3034 if (decoder->set_fup_mwait)
3035 no_tip = true;
3036 if (no_tip)
3037 decoder->pkt_state = INTEL_PT_STATE_FUP_NO_TIP;
3038 else
3039 decoder->pkt_state = INTEL_PT_STATE_FUP;
3040 err = intel_pt_walk_fup(decoder);
3041 if (err != -EAGAIN)
3042 return err;
3043 if (no_tip) {
3044 no_tip = false;
3045 break;
3046 }
3047 return intel_pt_walk_fup_tip(decoder);
3048
3049 case INTEL_PT_TRACESTOP:
3050 decoder->pge = false;
3051 decoder->continuous_period = false;
3052 intel_pt_clear_tx_flags(decoder);
3053 decoder->have_tma = false;
3054 break;
3055
3056 case INTEL_PT_PSB:
3057 decoder->state.psb_offset = decoder->pos;
3058 decoder->psb_ip = 0;
3059 if (intel_pt_psb_with_fup(decoder, &err))
3060 return err;
3061 err = intel_pt_psb(decoder);
3062 if (err == -EAGAIN)
3063 goto next;
3064 return err;
3065
3066 case INTEL_PT_PIP:
3067 intel_pt_update_pip(decoder);
3068 break;
3069
3070 case INTEL_PT_MTC:
3071 intel_pt_calc_mtc_timestamp(decoder);
3072 if (decoder->period_type != INTEL_PT_PERIOD_MTC)
3073 break;
3074 /*
3075 * Ensure that there has been an instruction since the
3076 * last MTC.
3077 */
3078 if (!decoder->mtc_insn)
3079 break;
3080 decoder->mtc_insn = false;
3081 /* Ensure that there is a timestamp */
3082 if (!decoder->timestamp)
3083 break;
3084 decoder->state.type = INTEL_PT_INSTRUCTION;
3085 decoder->state.from_ip = decoder->ip;
3086 decoder->state.to_ip = 0;
3087 decoder->mtc_insn = false;
3088 return 0;
3089
3090 case INTEL_PT_TSC:
3091 intel_pt_calc_tsc_timestamp(decoder);
3092 break;
3093
3094 case INTEL_PT_TMA:
3095 intel_pt_calc_tma(decoder);
3096 break;
3097
3098 case INTEL_PT_CYC:
3099 intel_pt_calc_cyc_timestamp(decoder);
3100 break;
3101
3102 case INTEL_PT_CBR:
3103 intel_pt_calc_cbr(decoder);
3104 if (decoder->cbr != decoder->cbr_seen) {
3105 decoder->state.type = 0;
3106 return 0;
3107 }
3108 break;
3109
3110 case INTEL_PT_MODE_EXEC:
3111 decoder->exec_mode = decoder->packet.payload;
3112 break;
3113
3114 case INTEL_PT_MODE_TSX:
3115 /* MODE_TSX need not be followed by FUP */
3116 if (!decoder->pge || decoder->in_psb) {
3117 intel_pt_update_in_tx(decoder);
3118 break;
3119 }
3120 err = intel_pt_mode_tsx(decoder, &no_tip);
3121 if (err)
3122 return err;
3123 goto next;
3124
3125 case INTEL_PT_BAD: /* Does not happen */
3126 return intel_pt_bug(decoder);
3127
3128 case INTEL_PT_PSBEND:
3129 case INTEL_PT_VMCS:
3130 case INTEL_PT_MNT:
3131 case INTEL_PT_PAD:
3132 break;
3133
3134 case INTEL_PT_PTWRITE_IP:
3135 decoder->fup_ptw_payload = decoder->packet.payload;
3136 err = intel_pt_get_next_packet(decoder);
3137 if (err)
3138 return err;
3139 if (decoder->packet.type == INTEL_PT_FUP) {
3140 decoder->set_fup_ptw = true;
3141 no_tip = true;
3142 } else {
3143 intel_pt_log_at("ERROR: Missing FUP after PTWRITE",
3144 decoder->pos);
3145 }
3146 goto next;
3147
3148 case INTEL_PT_PTWRITE:
3149 decoder->state.type = INTEL_PT_PTW;
3150 decoder->state.from_ip = decoder->ip;
3151 decoder->state.to_ip = 0;
3152 decoder->state.ptw_payload = decoder->packet.payload;
3153 return 0;
3154
3155 case INTEL_PT_MWAIT:
3156 decoder->fup_mwait_payload = decoder->packet.payload;
3157 decoder->set_fup_mwait = true;
3158 break;
3159
3160 case INTEL_PT_PWRE:
3161 if (decoder->set_fup_mwait) {
3162 decoder->fup_pwre_payload =
3163 decoder->packet.payload;
3164 decoder->set_fup_pwre = true;
3165 break;
3166 }
3167 decoder->state.type = INTEL_PT_PWR_ENTRY;
3168 decoder->state.from_ip = decoder->ip;
3169 decoder->state.to_ip = 0;
3170 decoder->state.pwrx_payload = decoder->packet.payload;
3171 return 0;
3172
3173 case INTEL_PT_EXSTOP_IP:
3174 err = intel_pt_get_next_packet(decoder);
3175 if (err)
3176 return err;
3177 if (decoder->packet.type == INTEL_PT_FUP) {
3178 decoder->set_fup_exstop = true;
3179 no_tip = true;
3180 } else {
3181 intel_pt_log_at("ERROR: Missing FUP after EXSTOP",
3182 decoder->pos);
3183 }
3184 goto next;
3185
3186 case INTEL_PT_EXSTOP:
3187 decoder->state.type = INTEL_PT_EX_STOP;
3188 decoder->state.from_ip = decoder->ip;
3189 decoder->state.to_ip = 0;
3190 return 0;
3191
3192 case INTEL_PT_PWRX:
3193 decoder->state.type = INTEL_PT_PWR_EXIT;
3194 decoder->state.from_ip = decoder->ip;
3195 decoder->state.to_ip = 0;
3196 decoder->state.pwrx_payload = decoder->packet.payload;
3197 return 0;
3198
3199 case INTEL_PT_BBP:
3200 intel_pt_bbp(decoder);
3201 break;
3202
3203 case INTEL_PT_BIP:
3204 intel_pt_bip(decoder);
3205 break;
3206
3207 case INTEL_PT_BEP:
3208 decoder->state.type = INTEL_PT_BLK_ITEMS;
3209 decoder->state.from_ip = decoder->ip;
3210 decoder->state.to_ip = 0;
3211 return 0;
3212
3213 case INTEL_PT_BEP_IP:
3214 err = intel_pt_get_next_packet(decoder);
3215 if (err)
3216 return err;
3217 if (decoder->packet.type == INTEL_PT_FUP) {
3218 decoder->set_fup_bep = true;
3219 no_tip = true;
3220 } else {
3221 intel_pt_log_at("ERROR: Missing FUP after BEP",
3222 decoder->pos);
3223 }
3224 goto next;
3225
3226 default:
3227 return intel_pt_bug(decoder);
3228 }
3229 }
3230 }
3231
intel_pt_have_ip(struct intel_pt_decoder * decoder)3232 static inline bool intel_pt_have_ip(struct intel_pt_decoder *decoder)
3233 {
3234 return decoder->packet.count &&
3235 (decoder->have_last_ip || decoder->packet.count == 3 ||
3236 decoder->packet.count == 6);
3237 }
3238
3239 /* Walk PSB+ packets to get in sync. */
intel_pt_walk_psb(struct intel_pt_decoder * decoder)3240 static int intel_pt_walk_psb(struct intel_pt_decoder *decoder)
3241 {
3242 int err;
3243
3244 decoder->in_psb = true;
3245
3246 while (1) {
3247 err = intel_pt_get_next_packet(decoder);
3248 if (err)
3249 goto out;
3250
3251 switch (decoder->packet.type) {
3252 case INTEL_PT_TIP_PGD:
3253 decoder->continuous_period = false;
3254 __fallthrough;
3255 case INTEL_PT_TIP_PGE:
3256 case INTEL_PT_TIP:
3257 case INTEL_PT_PTWRITE:
3258 case INTEL_PT_PTWRITE_IP:
3259 case INTEL_PT_EXSTOP:
3260 case INTEL_PT_EXSTOP_IP:
3261 case INTEL_PT_MWAIT:
3262 case INTEL_PT_PWRE:
3263 case INTEL_PT_PWRX:
3264 case INTEL_PT_BBP:
3265 case INTEL_PT_BIP:
3266 case INTEL_PT_BEP:
3267 case INTEL_PT_BEP_IP:
3268 intel_pt_log("ERROR: Unexpected packet\n");
3269 err = -ENOENT;
3270 goto out;
3271
3272 case INTEL_PT_FUP:
3273 decoder->pge = true;
3274 if (intel_pt_have_ip(decoder)) {
3275 uint64_t current_ip = decoder->ip;
3276
3277 intel_pt_set_ip(decoder);
3278 decoder->psb_ip = decoder->ip;
3279 if (current_ip)
3280 intel_pt_log_to("Setting IP",
3281 decoder->ip);
3282 }
3283 break;
3284
3285 case INTEL_PT_MTC:
3286 intel_pt_calc_mtc_timestamp(decoder);
3287 break;
3288
3289 case INTEL_PT_TSC:
3290 intel_pt_calc_tsc_timestamp(decoder);
3291 break;
3292
3293 case INTEL_PT_TMA:
3294 intel_pt_calc_tma(decoder);
3295 break;
3296
3297 case INTEL_PT_CYC:
3298 intel_pt_calc_cyc_timestamp(decoder);
3299 break;
3300
3301 case INTEL_PT_CBR:
3302 intel_pt_calc_cbr(decoder);
3303 break;
3304
3305 case INTEL_PT_PIP:
3306 intel_pt_set_pip(decoder);
3307 break;
3308
3309 case INTEL_PT_MODE_EXEC:
3310 decoder->exec_mode = decoder->packet.payload;
3311 break;
3312
3313 case INTEL_PT_MODE_TSX:
3314 intel_pt_update_in_tx(decoder);
3315 break;
3316
3317 case INTEL_PT_TRACESTOP:
3318 decoder->pge = false;
3319 decoder->continuous_period = false;
3320 intel_pt_clear_tx_flags(decoder);
3321 __fallthrough;
3322
3323 case INTEL_PT_TNT:
3324 decoder->have_tma = false;
3325 intel_pt_log("ERROR: Unexpected packet\n");
3326 if (decoder->ip)
3327 decoder->pkt_state = INTEL_PT_STATE_ERR4;
3328 else
3329 decoder->pkt_state = INTEL_PT_STATE_ERR3;
3330 err = -ENOENT;
3331 goto out;
3332
3333 case INTEL_PT_BAD: /* Does not happen */
3334 err = intel_pt_bug(decoder);
3335 goto out;
3336
3337 case INTEL_PT_OVF:
3338 err = intel_pt_overflow(decoder);
3339 goto out;
3340
3341 case INTEL_PT_PSBEND:
3342 err = 0;
3343 goto out;
3344
3345 case INTEL_PT_PSB:
3346 case INTEL_PT_VMCS:
3347 case INTEL_PT_MNT:
3348 case INTEL_PT_PAD:
3349 default:
3350 break;
3351 }
3352 }
3353 out:
3354 decoder->in_psb = false;
3355
3356 return err;
3357 }
3358
intel_pt_walk_to_ip(struct intel_pt_decoder * decoder)3359 static int intel_pt_walk_to_ip(struct intel_pt_decoder *decoder)
3360 {
3361 int err;
3362
3363 while (1) {
3364 err = intel_pt_get_next_packet(decoder);
3365 if (err)
3366 return err;
3367
3368 switch (decoder->packet.type) {
3369 case INTEL_PT_TIP_PGD:
3370 decoder->continuous_period = false;
3371 decoder->pge = false;
3372 if (intel_pt_have_ip(decoder))
3373 intel_pt_set_ip(decoder);
3374 if (!decoder->ip)
3375 break;
3376 decoder->state.type |= INTEL_PT_TRACE_END;
3377 return 0;
3378
3379 case INTEL_PT_TIP_PGE:
3380 decoder->pge = true;
3381 intel_pt_mtc_cyc_cnt_pge(decoder);
3382 if (intel_pt_have_ip(decoder))
3383 intel_pt_set_ip(decoder);
3384 if (!decoder->ip)
3385 break;
3386 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
3387 return 0;
3388
3389 case INTEL_PT_TIP:
3390 decoder->pge = true;
3391 if (intel_pt_have_ip(decoder))
3392 intel_pt_set_ip(decoder);
3393 if (!decoder->ip)
3394 break;
3395 return 0;
3396
3397 case INTEL_PT_FUP:
3398 if (intel_pt_have_ip(decoder))
3399 intel_pt_set_ip(decoder);
3400 if (decoder->ip)
3401 return 0;
3402 break;
3403
3404 case INTEL_PT_MTC:
3405 intel_pt_calc_mtc_timestamp(decoder);
3406 break;
3407
3408 case INTEL_PT_TSC:
3409 intel_pt_calc_tsc_timestamp(decoder);
3410 break;
3411
3412 case INTEL_PT_TMA:
3413 intel_pt_calc_tma(decoder);
3414 break;
3415
3416 case INTEL_PT_CYC:
3417 intel_pt_calc_cyc_timestamp(decoder);
3418 break;
3419
3420 case INTEL_PT_CBR:
3421 intel_pt_calc_cbr(decoder);
3422 break;
3423
3424 case INTEL_PT_PIP:
3425 intel_pt_set_pip(decoder);
3426 break;
3427
3428 case INTEL_PT_MODE_EXEC:
3429 decoder->exec_mode = decoder->packet.payload;
3430 break;
3431
3432 case INTEL_PT_MODE_TSX:
3433 intel_pt_update_in_tx(decoder);
3434 break;
3435
3436 case INTEL_PT_OVF:
3437 return intel_pt_overflow(decoder);
3438
3439 case INTEL_PT_BAD: /* Does not happen */
3440 return intel_pt_bug(decoder);
3441
3442 case INTEL_PT_TRACESTOP:
3443 decoder->pge = false;
3444 decoder->continuous_period = false;
3445 intel_pt_clear_tx_flags(decoder);
3446 decoder->have_tma = false;
3447 break;
3448
3449 case INTEL_PT_PSB:
3450 decoder->state.psb_offset = decoder->pos;
3451 decoder->psb_ip = 0;
3452 decoder->last_ip = 0;
3453 decoder->have_last_ip = true;
3454 intel_pt_clear_stack(&decoder->stack);
3455 err = intel_pt_walk_psb(decoder);
3456 if (err)
3457 return err;
3458 decoder->state.type = INTEL_PT_PSB_EVT;
3459 decoder->state.from_ip = decoder->psb_ip;
3460 decoder->state.to_ip = 0;
3461 return 0;
3462
3463 case INTEL_PT_TNT:
3464 case INTEL_PT_PSBEND:
3465 case INTEL_PT_VMCS:
3466 case INTEL_PT_MNT:
3467 case INTEL_PT_PAD:
3468 case INTEL_PT_PTWRITE:
3469 case INTEL_PT_PTWRITE_IP:
3470 case INTEL_PT_EXSTOP:
3471 case INTEL_PT_EXSTOP_IP:
3472 case INTEL_PT_MWAIT:
3473 case INTEL_PT_PWRE:
3474 case INTEL_PT_PWRX:
3475 case INTEL_PT_BBP:
3476 case INTEL_PT_BIP:
3477 case INTEL_PT_BEP:
3478 case INTEL_PT_BEP_IP:
3479 default:
3480 break;
3481 }
3482 }
3483 }
3484
intel_pt_sync_ip(struct intel_pt_decoder * decoder)3485 static int intel_pt_sync_ip(struct intel_pt_decoder *decoder)
3486 {
3487 int err;
3488
3489 decoder->set_fup_tx_flags = false;
3490 decoder->set_fup_ptw = false;
3491 decoder->set_fup_mwait = false;
3492 decoder->set_fup_pwre = false;
3493 decoder->set_fup_exstop = false;
3494 decoder->set_fup_bep = false;
3495 decoder->overflow = false;
3496
3497 if (!decoder->branch_enable) {
3498 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3499 decoder->state.type = 0; /* Do not have a sample */
3500 return 0;
3501 }
3502
3503 intel_pt_log("Scanning for full IP\n");
3504 err = intel_pt_walk_to_ip(decoder);
3505 if (err || ((decoder->state.type & INTEL_PT_PSB_EVT) && !decoder->ip))
3506 return err;
3507
3508 /* In hop mode, resample to get the to_ip as an "instruction" sample */
3509 if (decoder->hop)
3510 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
3511 else
3512 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3513
3514 decoder->state.from_ip = 0;
3515 decoder->state.to_ip = decoder->ip;
3516 intel_pt_log_to("Setting IP", decoder->ip);
3517
3518 return 0;
3519 }
3520
intel_pt_part_psb(struct intel_pt_decoder * decoder)3521 static int intel_pt_part_psb(struct intel_pt_decoder *decoder)
3522 {
3523 const unsigned char *end = decoder->buf + decoder->len;
3524 size_t i;
3525
3526 for (i = INTEL_PT_PSB_LEN - 1; i; i--) {
3527 if (i > decoder->len)
3528 continue;
3529 if (!memcmp(end - i, INTEL_PT_PSB_STR, i))
3530 return i;
3531 }
3532 return 0;
3533 }
3534
intel_pt_rest_psb(struct intel_pt_decoder * decoder,int part_psb)3535 static int intel_pt_rest_psb(struct intel_pt_decoder *decoder, int part_psb)
3536 {
3537 size_t rest_psb = INTEL_PT_PSB_LEN - part_psb;
3538 const char *psb = INTEL_PT_PSB_STR;
3539
3540 if (rest_psb > decoder->len ||
3541 memcmp(decoder->buf, psb + part_psb, rest_psb))
3542 return 0;
3543
3544 return rest_psb;
3545 }
3546
intel_pt_get_split_psb(struct intel_pt_decoder * decoder,int part_psb)3547 static int intel_pt_get_split_psb(struct intel_pt_decoder *decoder,
3548 int part_psb)
3549 {
3550 int rest_psb, ret;
3551
3552 decoder->pos += decoder->len;
3553 decoder->len = 0;
3554
3555 ret = intel_pt_get_next_data(decoder, false);
3556 if (ret)
3557 return ret;
3558
3559 rest_psb = intel_pt_rest_psb(decoder, part_psb);
3560 if (!rest_psb)
3561 return 0;
3562
3563 decoder->pos -= part_psb;
3564 decoder->next_buf = decoder->buf + rest_psb;
3565 decoder->next_len = decoder->len - rest_psb;
3566 memcpy(decoder->temp_buf, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
3567 decoder->buf = decoder->temp_buf;
3568 decoder->len = INTEL_PT_PSB_LEN;
3569
3570 return 0;
3571 }
3572
intel_pt_scan_for_psb(struct intel_pt_decoder * decoder)3573 static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder)
3574 {
3575 unsigned char *next;
3576 int ret;
3577
3578 intel_pt_log("Scanning for PSB\n");
3579 while (1) {
3580 if (!decoder->len) {
3581 ret = intel_pt_get_next_data(decoder, false);
3582 if (ret)
3583 return ret;
3584 }
3585
3586 next = memmem(decoder->buf, decoder->len, INTEL_PT_PSB_STR,
3587 INTEL_PT_PSB_LEN);
3588 if (!next) {
3589 int part_psb;
3590
3591 part_psb = intel_pt_part_psb(decoder);
3592 if (part_psb) {
3593 ret = intel_pt_get_split_psb(decoder, part_psb);
3594 if (ret)
3595 return ret;
3596 } else {
3597 decoder->pos += decoder->len;
3598 decoder->len = 0;
3599 }
3600 continue;
3601 }
3602
3603 decoder->pkt_step = next - decoder->buf;
3604 return intel_pt_get_next_packet(decoder);
3605 }
3606 }
3607
intel_pt_sync(struct intel_pt_decoder * decoder)3608 static int intel_pt_sync(struct intel_pt_decoder *decoder)
3609 {
3610 int err;
3611
3612 decoder->pge = false;
3613 decoder->continuous_period = false;
3614 decoder->have_last_ip = false;
3615 decoder->last_ip = 0;
3616 decoder->psb_ip = 0;
3617 decoder->ip = 0;
3618 intel_pt_clear_stack(&decoder->stack);
3619
3620 err = intel_pt_scan_for_psb(decoder);
3621 if (err)
3622 return err;
3623
3624 if (decoder->vm_time_correlation) {
3625 decoder->in_psb = true;
3626 if (!decoder->timestamp)
3627 decoder->timestamp = 1;
3628 decoder->state.type = 0;
3629 decoder->pkt_state = INTEL_PT_STATE_VM_TIME_CORRELATION;
3630 return 0;
3631 }
3632
3633 decoder->have_last_ip = true;
3634 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3635
3636 err = intel_pt_walk_psb(decoder);
3637 if (err)
3638 return err;
3639
3640 decoder->state.type = INTEL_PT_PSB_EVT; /* Only PSB sample */
3641 decoder->state.from_ip = decoder->psb_ip;
3642 decoder->state.to_ip = 0;
3643
3644 if (decoder->ip) {
3645 /*
3646 * In hop mode, resample to get the PSB FUP ip as an
3647 * "instruction" sample.
3648 */
3649 if (decoder->hop)
3650 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
3651 else
3652 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3653 }
3654
3655 return 0;
3656 }
3657
intel_pt_est_timestamp(struct intel_pt_decoder * decoder)3658 static uint64_t intel_pt_est_timestamp(struct intel_pt_decoder *decoder)
3659 {
3660 uint64_t est = decoder->sample_insn_cnt << 1;
3661
3662 if (!decoder->cbr || !decoder->max_non_turbo_ratio)
3663 goto out;
3664
3665 est *= decoder->max_non_turbo_ratio;
3666 est /= decoder->cbr;
3667 out:
3668 return decoder->sample_timestamp + est;
3669 }
3670
intel_pt_decode(struct intel_pt_decoder * decoder)3671 const struct intel_pt_state *intel_pt_decode(struct intel_pt_decoder *decoder)
3672 {
3673 int err;
3674
3675 do {
3676 decoder->state.type = INTEL_PT_BRANCH;
3677 decoder->state.flags = 0;
3678
3679 switch (decoder->pkt_state) {
3680 case INTEL_PT_STATE_NO_PSB:
3681 err = intel_pt_sync(decoder);
3682 break;
3683 case INTEL_PT_STATE_NO_IP:
3684 decoder->have_last_ip = false;
3685 decoder->last_ip = 0;
3686 decoder->ip = 0;
3687 __fallthrough;
3688 case INTEL_PT_STATE_ERR_RESYNC:
3689 err = intel_pt_sync_ip(decoder);
3690 break;
3691 case INTEL_PT_STATE_IN_SYNC:
3692 err = intel_pt_walk_trace(decoder);
3693 break;
3694 case INTEL_PT_STATE_TNT:
3695 case INTEL_PT_STATE_TNT_CONT:
3696 err = intel_pt_walk_tnt(decoder);
3697 if (err == -EAGAIN)
3698 err = intel_pt_walk_trace(decoder);
3699 break;
3700 case INTEL_PT_STATE_TIP:
3701 case INTEL_PT_STATE_TIP_PGD:
3702 err = intel_pt_walk_tip(decoder);
3703 break;
3704 case INTEL_PT_STATE_FUP:
3705 err = intel_pt_walk_fup(decoder);
3706 if (err == -EAGAIN)
3707 err = intel_pt_walk_fup_tip(decoder);
3708 break;
3709 case INTEL_PT_STATE_FUP_NO_TIP:
3710 err = intel_pt_walk_fup(decoder);
3711 if (err == -EAGAIN)
3712 err = intel_pt_walk_trace(decoder);
3713 break;
3714 case INTEL_PT_STATE_FUP_IN_PSB:
3715 err = intel_pt_fup_in_psb(decoder);
3716 break;
3717 case INTEL_PT_STATE_RESAMPLE:
3718 err = intel_pt_resample(decoder);
3719 break;
3720 case INTEL_PT_STATE_VM_TIME_CORRELATION:
3721 err = intel_pt_vm_time_correlation(decoder);
3722 break;
3723 default:
3724 err = intel_pt_bug(decoder);
3725 break;
3726 }
3727 } while (err == -ENOLINK);
3728
3729 if (err) {
3730 decoder->state.err = intel_pt_ext_err(err);
3731 if (err != -EOVERFLOW)
3732 decoder->state.from_ip = decoder->ip;
3733 intel_pt_update_sample_time(decoder);
3734 decoder->sample_tot_cyc_cnt = decoder->tot_cyc_cnt;
3735 intel_pt_set_nr(decoder);
3736 } else {
3737 decoder->state.err = 0;
3738 if (decoder->cbr != decoder->cbr_seen) {
3739 decoder->cbr_seen = decoder->cbr;
3740 if (!decoder->state.type) {
3741 decoder->state.from_ip = decoder->ip;
3742 decoder->state.to_ip = 0;
3743 }
3744 decoder->state.type |= INTEL_PT_CBR_CHG;
3745 decoder->state.cbr_payload = decoder->cbr_payload;
3746 decoder->state.cbr = decoder->cbr;
3747 }
3748 if (intel_pt_sample_time(decoder->pkt_state)) {
3749 intel_pt_update_sample_time(decoder);
3750 if (decoder->sample_cyc) {
3751 decoder->sample_tot_cyc_cnt = decoder->tot_cyc_cnt;
3752 decoder->state.flags |= INTEL_PT_SAMPLE_IPC;
3753 decoder->sample_cyc = false;
3754 }
3755 }
3756 /*
3757 * When using only TSC/MTC to compute cycles, IPC can be
3758 * sampled as soon as the cycle count changes.
3759 */
3760 if (!decoder->have_cyc)
3761 decoder->state.flags |= INTEL_PT_SAMPLE_IPC;
3762 }
3763
3764 /* Let PSB event always have TSC timestamp */
3765 if ((decoder->state.type & INTEL_PT_PSB_EVT) && decoder->tsc_timestamp)
3766 decoder->sample_timestamp = decoder->tsc_timestamp;
3767
3768 decoder->state.from_nr = decoder->nr;
3769 decoder->state.to_nr = decoder->next_nr;
3770 decoder->nr = decoder->next_nr;
3771
3772 decoder->state.timestamp = decoder->sample_timestamp;
3773 decoder->state.est_timestamp = intel_pt_est_timestamp(decoder);
3774 decoder->state.tot_insn_cnt = decoder->tot_insn_cnt;
3775 decoder->state.tot_cyc_cnt = decoder->sample_tot_cyc_cnt;
3776
3777 return &decoder->state;
3778 }
3779
3780 /**
3781 * intel_pt_next_psb - move buffer pointer to the start of the next PSB packet.
3782 * @buf: pointer to buffer pointer
3783 * @len: size of buffer
3784 *
3785 * Updates the buffer pointer to point to the start of the next PSB packet if
3786 * there is one, otherwise the buffer pointer is unchanged. If @buf is updated,
3787 * @len is adjusted accordingly.
3788 *
3789 * Return: %true if a PSB packet is found, %false otherwise.
3790 */
intel_pt_next_psb(unsigned char ** buf,size_t * len)3791 static bool intel_pt_next_psb(unsigned char **buf, size_t *len)
3792 {
3793 unsigned char *next;
3794
3795 next = memmem(*buf, *len, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
3796 if (next) {
3797 *len -= next - *buf;
3798 *buf = next;
3799 return true;
3800 }
3801 return false;
3802 }
3803
3804 /**
3805 * intel_pt_step_psb - move buffer pointer to the start of the following PSB
3806 * packet.
3807 * @buf: pointer to buffer pointer
3808 * @len: size of buffer
3809 *
3810 * Updates the buffer pointer to point to the start of the following PSB packet
3811 * (skipping the PSB at @buf itself) if there is one, otherwise the buffer
3812 * pointer is unchanged. If @buf is updated, @len is adjusted accordingly.
3813 *
3814 * Return: %true if a PSB packet is found, %false otherwise.
3815 */
intel_pt_step_psb(unsigned char ** buf,size_t * len)3816 static bool intel_pt_step_psb(unsigned char **buf, size_t *len)
3817 {
3818 unsigned char *next;
3819
3820 if (!*len)
3821 return false;
3822
3823 next = memmem(*buf + 1, *len - 1, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
3824 if (next) {
3825 *len -= next - *buf;
3826 *buf = next;
3827 return true;
3828 }
3829 return false;
3830 }
3831
3832 /**
3833 * intel_pt_last_psb - find the last PSB packet in a buffer.
3834 * @buf: buffer
3835 * @len: size of buffer
3836 *
3837 * This function finds the last PSB in a buffer.
3838 *
3839 * Return: A pointer to the last PSB in @buf if found, %NULL otherwise.
3840 */
intel_pt_last_psb(unsigned char * buf,size_t len)3841 static unsigned char *intel_pt_last_psb(unsigned char *buf, size_t len)
3842 {
3843 const char *n = INTEL_PT_PSB_STR;
3844 unsigned char *p;
3845 size_t k;
3846
3847 if (len < INTEL_PT_PSB_LEN)
3848 return NULL;
3849
3850 k = len - INTEL_PT_PSB_LEN + 1;
3851 while (1) {
3852 p = memrchr(buf, n[0], k);
3853 if (!p)
3854 return NULL;
3855 if (!memcmp(p + 1, n + 1, INTEL_PT_PSB_LEN - 1))
3856 return p;
3857 k = p - buf;
3858 if (!k)
3859 return NULL;
3860 }
3861 }
3862
3863 /**
3864 * intel_pt_next_tsc - find and return next TSC.
3865 * @buf: buffer
3866 * @len: size of buffer
3867 * @tsc: TSC value returned
3868 * @rem: returns remaining size when TSC is found
3869 *
3870 * Find a TSC packet in @buf and return the TSC value. This function assumes
3871 * that @buf starts at a PSB and that PSB+ will contain TSC and so stops if a
3872 * PSBEND packet is found.
3873 *
3874 * Return: %true if TSC is found, false otherwise.
3875 */
intel_pt_next_tsc(unsigned char * buf,size_t len,uint64_t * tsc,size_t * rem)3876 static bool intel_pt_next_tsc(unsigned char *buf, size_t len, uint64_t *tsc,
3877 size_t *rem)
3878 {
3879 enum intel_pt_pkt_ctx ctx = INTEL_PT_NO_CTX;
3880 struct intel_pt_pkt packet;
3881 int ret;
3882
3883 while (len) {
3884 ret = intel_pt_get_packet(buf, len, &packet, &ctx);
3885 if (ret <= 0)
3886 return false;
3887 if (packet.type == INTEL_PT_TSC) {
3888 *tsc = packet.payload;
3889 *rem = len;
3890 return true;
3891 }
3892 if (packet.type == INTEL_PT_PSBEND)
3893 return false;
3894 buf += ret;
3895 len -= ret;
3896 }
3897 return false;
3898 }
3899
3900 /**
3901 * intel_pt_tsc_cmp - compare 7-byte TSCs.
3902 * @tsc1: first TSC to compare
3903 * @tsc2: second TSC to compare
3904 *
3905 * This function compares 7-byte TSC values allowing for the possibility that
3906 * TSC wrapped around. Generally it is not possible to know if TSC has wrapped
3907 * around so for that purpose this function assumes the absolute difference is
3908 * less than half the maximum difference.
3909 *
3910 * Return: %-1 if @tsc1 is before @tsc2, %0 if @tsc1 == @tsc2, %1 if @tsc1 is
3911 * after @tsc2.
3912 */
intel_pt_tsc_cmp(uint64_t tsc1,uint64_t tsc2)3913 static int intel_pt_tsc_cmp(uint64_t tsc1, uint64_t tsc2)
3914 {
3915 const uint64_t halfway = (1ULL << 55);
3916
3917 if (tsc1 == tsc2)
3918 return 0;
3919
3920 if (tsc1 < tsc2) {
3921 if (tsc2 - tsc1 < halfway)
3922 return -1;
3923 else
3924 return 1;
3925 } else {
3926 if (tsc1 - tsc2 < halfway)
3927 return 1;
3928 else
3929 return -1;
3930 }
3931 }
3932
3933 #define MAX_PADDING (PERF_AUXTRACE_RECORD_ALIGNMENT - 1)
3934
3935 /**
3936 * adj_for_padding - adjust overlap to account for padding.
3937 * @buf_b: second buffer
3938 * @buf_a: first buffer
3939 * @len_a: size of first buffer
3940 *
3941 * @buf_a might have up to 7 bytes of padding appended. Adjust the overlap
3942 * accordingly.
3943 *
3944 * Return: A pointer into @buf_b from where non-overlapped data starts
3945 */
adj_for_padding(unsigned char * buf_b,unsigned char * buf_a,size_t len_a)3946 static unsigned char *adj_for_padding(unsigned char *buf_b,
3947 unsigned char *buf_a, size_t len_a)
3948 {
3949 unsigned char *p = buf_b - MAX_PADDING;
3950 unsigned char *q = buf_a + len_a - MAX_PADDING;
3951 int i;
3952
3953 for (i = MAX_PADDING; i; i--, p++, q++) {
3954 if (*p != *q)
3955 break;
3956 }
3957
3958 return p;
3959 }
3960
3961 /**
3962 * intel_pt_find_overlap_tsc - determine start of non-overlapped trace data
3963 * using TSC.
3964 * @buf_a: first buffer
3965 * @len_a: size of first buffer
3966 * @buf_b: second buffer
3967 * @len_b: size of second buffer
3968 * @consecutive: returns true if there is data in buf_b that is consecutive
3969 * to buf_a
3970 * @ooo_tsc: out-of-order TSC due to VM TSC offset / scaling
3971 *
3972 * If the trace contains TSC we can look at the last TSC of @buf_a and the
3973 * first TSC of @buf_b in order to determine if the buffers overlap, and then
3974 * walk forward in @buf_b until a later TSC is found. A precondition is that
3975 * @buf_a and @buf_b are positioned at a PSB.
3976 *
3977 * Return: A pointer into @buf_b from where non-overlapped data starts, or
3978 * @buf_b + @len_b if there is no non-overlapped data.
3979 */
intel_pt_find_overlap_tsc(unsigned char * buf_a,size_t len_a,unsigned char * buf_b,size_t len_b,bool * consecutive,bool ooo_tsc)3980 static unsigned char *intel_pt_find_overlap_tsc(unsigned char *buf_a,
3981 size_t len_a,
3982 unsigned char *buf_b,
3983 size_t len_b, bool *consecutive,
3984 bool ooo_tsc)
3985 {
3986 uint64_t tsc_a, tsc_b;
3987 unsigned char *p;
3988 size_t len, rem_a, rem_b;
3989
3990 p = intel_pt_last_psb(buf_a, len_a);
3991 if (!p)
3992 return buf_b; /* No PSB in buf_a => no overlap */
3993
3994 len = len_a - (p - buf_a);
3995 if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a)) {
3996 /* The last PSB+ in buf_a is incomplete, so go back one more */
3997 len_a -= len;
3998 p = intel_pt_last_psb(buf_a, len_a);
3999 if (!p)
4000 return buf_b; /* No full PSB+ => assume no overlap */
4001 len = len_a - (p - buf_a);
4002 if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a))
4003 return buf_b; /* No TSC in buf_a => assume no overlap */
4004 }
4005
4006 while (1) {
4007 /* Ignore PSB+ with no TSC */
4008 if (intel_pt_next_tsc(buf_b, len_b, &tsc_b, &rem_b)) {
4009 int cmp = intel_pt_tsc_cmp(tsc_a, tsc_b);
4010
4011 /* Same TSC, so buffers are consecutive */
4012 if (!cmp && rem_b >= rem_a) {
4013 unsigned char *start;
4014
4015 *consecutive = true;
4016 start = buf_b + len_b - (rem_b - rem_a);
4017 return adj_for_padding(start, buf_a, len_a);
4018 }
4019 if (cmp < 0 && !ooo_tsc)
4020 return buf_b; /* tsc_a < tsc_b => no overlap */
4021 }
4022
4023 if (!intel_pt_step_psb(&buf_b, &len_b))
4024 return buf_b + len_b; /* No PSB in buf_b => no data */
4025 }
4026 }
4027
4028 /**
4029 * intel_pt_find_overlap - determine start of non-overlapped trace data.
4030 * @buf_a: first buffer
4031 * @len_a: size of first buffer
4032 * @buf_b: second buffer
4033 * @len_b: size of second buffer
4034 * @have_tsc: can use TSC packets to detect overlap
4035 * @consecutive: returns true if there is data in buf_b that is consecutive
4036 * to buf_a
4037 * @ooo_tsc: out-of-order TSC due to VM TSC offset / scaling
4038 *
4039 * When trace samples or snapshots are recorded there is the possibility that
4040 * the data overlaps. Note that, for the purposes of decoding, data is only
4041 * useful if it begins with a PSB packet.
4042 *
4043 * Return: A pointer into @buf_b from where non-overlapped data starts, or
4044 * @buf_b + @len_b if there is no non-overlapped data.
4045 */
intel_pt_find_overlap(unsigned char * buf_a,size_t len_a,unsigned char * buf_b,size_t len_b,bool have_tsc,bool * consecutive,bool ooo_tsc)4046 unsigned char *intel_pt_find_overlap(unsigned char *buf_a, size_t len_a,
4047 unsigned char *buf_b, size_t len_b,
4048 bool have_tsc, bool *consecutive,
4049 bool ooo_tsc)
4050 {
4051 unsigned char *found;
4052
4053 /* Buffer 'b' must start at PSB so throw away everything before that */
4054 if (!intel_pt_next_psb(&buf_b, &len_b))
4055 return buf_b + len_b; /* No PSB */
4056
4057 if (!intel_pt_next_psb(&buf_a, &len_a))
4058 return buf_b; /* No overlap */
4059
4060 if (have_tsc) {
4061 found = intel_pt_find_overlap_tsc(buf_a, len_a, buf_b, len_b,
4062 consecutive, ooo_tsc);
4063 if (found)
4064 return found;
4065 }
4066
4067 /*
4068 * Buffer 'b' cannot end within buffer 'a' so, for comparison purposes,
4069 * we can ignore the first part of buffer 'a'.
4070 */
4071 while (len_b < len_a) {
4072 if (!intel_pt_step_psb(&buf_a, &len_a))
4073 return buf_b; /* No overlap */
4074 }
4075
4076 /* Now len_b >= len_a */
4077 while (1) {
4078 /* Potential overlap so check the bytes */
4079 found = memmem(buf_a, len_a, buf_b, len_a);
4080 if (found) {
4081 *consecutive = true;
4082 return adj_for_padding(buf_b + len_a, buf_a, len_a);
4083 }
4084
4085 /* Try again at next PSB in buffer 'a' */
4086 if (!intel_pt_step_psb(&buf_a, &len_a))
4087 return buf_b; /* No overlap */
4088 }
4089 }
4090
4091 /**
4092 * struct fast_forward_data - data used by intel_pt_ff_cb().
4093 * @timestamp: timestamp to fast forward towards
4094 * @buf_timestamp: buffer timestamp of last buffer with trace data earlier than
4095 * the fast forward timestamp.
4096 */
4097 struct fast_forward_data {
4098 uint64_t timestamp;
4099 uint64_t buf_timestamp;
4100 };
4101
4102 /**
4103 * intel_pt_ff_cb - fast forward lookahead callback.
4104 * @buffer: Intel PT trace buffer
4105 * @data: opaque pointer to fast forward data (struct fast_forward_data)
4106 *
4107 * Determine if @buffer trace is past the fast forward timestamp.
4108 *
4109 * Return: 1 (stop lookahead) if @buffer trace is past the fast forward
4110 * timestamp, and 0 otherwise.
4111 */
intel_pt_ff_cb(struct intel_pt_buffer * buffer,void * data)4112 static int intel_pt_ff_cb(struct intel_pt_buffer *buffer, void *data)
4113 {
4114 struct fast_forward_data *d = data;
4115 unsigned char *buf;
4116 uint64_t tsc;
4117 size_t rem;
4118 size_t len;
4119
4120 buf = (unsigned char *)buffer->buf;
4121 len = buffer->len;
4122
4123 if (!intel_pt_next_psb(&buf, &len) ||
4124 !intel_pt_next_tsc(buf, len, &tsc, &rem))
4125 return 0;
4126
4127 tsc = intel_pt_8b_tsc(tsc, buffer->ref_timestamp);
4128
4129 intel_pt_log("Buffer 1st timestamp " x64_fmt " ref timestamp " x64_fmt "\n",
4130 tsc, buffer->ref_timestamp);
4131
4132 /*
4133 * If the buffer contains a timestamp earlier that the fast forward
4134 * timestamp, then record it, else stop.
4135 */
4136 if (tsc < d->timestamp)
4137 d->buf_timestamp = buffer->ref_timestamp;
4138 else
4139 return 1;
4140
4141 return 0;
4142 }
4143
4144 /**
4145 * intel_pt_fast_forward - reposition decoder forwards.
4146 * @decoder: Intel PT decoder
4147 * @timestamp: timestamp to fast forward towards
4148 *
4149 * Reposition decoder at the last PSB with a timestamp earlier than @timestamp.
4150 *
4151 * Return: 0 on success or negative error code on failure.
4152 */
intel_pt_fast_forward(struct intel_pt_decoder * decoder,uint64_t timestamp)4153 int intel_pt_fast_forward(struct intel_pt_decoder *decoder, uint64_t timestamp)
4154 {
4155 struct fast_forward_data d = { .timestamp = timestamp };
4156 unsigned char *buf;
4157 size_t len;
4158 int err;
4159
4160 intel_pt_log("Fast forward towards timestamp " x64_fmt "\n", timestamp);
4161
4162 /* Find buffer timestamp of buffer to fast forward to */
4163 err = decoder->lookahead(decoder->data, intel_pt_ff_cb, &d);
4164 if (err < 0)
4165 return err;
4166
4167 /* Walk to buffer with same buffer timestamp */
4168 if (d.buf_timestamp) {
4169 do {
4170 decoder->pos += decoder->len;
4171 decoder->len = 0;
4172 err = intel_pt_get_next_data(decoder, true);
4173 /* -ENOLINK means non-consecutive trace */
4174 if (err && err != -ENOLINK)
4175 return err;
4176 } while (decoder->buf_timestamp != d.buf_timestamp);
4177 }
4178
4179 if (!decoder->buf)
4180 return 0;
4181
4182 buf = (unsigned char *)decoder->buf;
4183 len = decoder->len;
4184
4185 if (!intel_pt_next_psb(&buf, &len))
4186 return 0;
4187
4188 /*
4189 * Walk PSBs while the PSB timestamp is less than the fast forward
4190 * timestamp.
4191 */
4192 do {
4193 uint64_t tsc;
4194 size_t rem;
4195
4196 if (!intel_pt_next_tsc(buf, len, &tsc, &rem))
4197 break;
4198 tsc = intel_pt_8b_tsc(tsc, decoder->buf_timestamp);
4199 /*
4200 * A TSC packet can slip past MTC packets but, after fast
4201 * forward, decoding starts at the TSC timestamp. That means
4202 * the timestamps may not be exactly the same as the timestamps
4203 * that would have been decoded without fast forward.
4204 */
4205 if (tsc < timestamp) {
4206 intel_pt_log("Fast forward to next PSB timestamp " x64_fmt "\n", tsc);
4207 decoder->pos += decoder->len - len;
4208 decoder->buf = buf;
4209 decoder->len = len;
4210 intel_pt_reposition(decoder);
4211 } else {
4212 break;
4213 }
4214 } while (intel_pt_step_psb(&buf, &len));
4215
4216 return 0;
4217 }
4218