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 emulated_ptwrite;
141 bool vm_time_correlation;
142 bool vm_tm_corr_dry_run;
143 bool vm_tm_corr_reliable;
144 bool vm_tm_corr_same_buf;
145 bool vm_tm_corr_continuous;
146 bool nr;
147 bool next_nr;
148 enum intel_pt_param_flags flags;
149 uint64_t pos;
150 uint64_t last_ip;
151 uint64_t ip;
152 uint64_t pip_payload;
153 uint64_t timestamp;
154 uint64_t tsc_timestamp;
155 uint64_t ref_timestamp;
156 uint64_t buf_timestamp;
157 uint64_t sample_timestamp;
158 uint64_t ret_addr;
159 uint64_t ctc_timestamp;
160 uint64_t ctc_delta;
161 uint64_t cycle_cnt;
162 uint64_t cyc_ref_timestamp;
163 uint64_t first_timestamp;
164 uint64_t last_reliable_timestamp;
165 uint64_t vmcs;
166 uint64_t print_once;
167 uint64_t last_ctc;
168 uint32_t last_mtc;
169 uint32_t tsc_ctc_ratio_n;
170 uint32_t tsc_ctc_ratio_d;
171 uint32_t tsc_ctc_mult;
172 uint32_t tsc_slip;
173 uint32_t ctc_rem_mask;
174 int mtc_shift;
175 struct intel_pt_stack stack;
176 enum intel_pt_pkt_state pkt_state;
177 enum intel_pt_pkt_ctx pkt_ctx;
178 enum intel_pt_pkt_ctx prev_pkt_ctx;
179 enum intel_pt_blk_type blk_type;
180 int blk_type_pos;
181 struct intel_pt_pkt packet;
182 struct intel_pt_pkt tnt;
183 int pkt_step;
184 int pkt_len;
185 int last_packet_type;
186 unsigned int cbr;
187 unsigned int cbr_seen;
188 unsigned int max_non_turbo_ratio;
189 double max_non_turbo_ratio_fp;
190 double cbr_cyc_to_tsc;
191 double calc_cyc_to_tsc;
192 bool have_calc_cyc_to_tsc;
193 int exec_mode;
194 unsigned int insn_bytes;
195 uint64_t period;
196 enum intel_pt_period_type period_type;
197 uint64_t tot_insn_cnt;
198 uint64_t period_insn_cnt;
199 uint64_t period_mask;
200 uint64_t period_ticks;
201 uint64_t last_masked_timestamp;
202 uint64_t tot_cyc_cnt;
203 uint64_t sample_tot_cyc_cnt;
204 uint64_t base_cyc_cnt;
205 uint64_t cyc_cnt_timestamp;
206 uint64_t ctl;
207 uint64_t cyc_threshold;
208 double tsc_to_cyc;
209 bool continuous_period;
210 bool overflow;
211 bool set_fup_tx_flags;
212 bool set_fup_ptw;
213 bool set_fup_mwait;
214 bool set_fup_pwre;
215 bool set_fup_exstop;
216 bool set_fup_bep;
217 bool sample_cyc;
218 unsigned int fup_tx_flags;
219 unsigned int tx_flags;
220 uint64_t fup_ptw_payload;
221 uint64_t fup_mwait_payload;
222 uint64_t fup_pwre_payload;
223 uint64_t cbr_payload;
224 uint64_t timestamp_insn_cnt;
225 uint64_t sample_insn_cnt;
226 uint64_t stuck_ip;
227 int max_loops;
228 int no_progress;
229 int stuck_ip_prd;
230 int stuck_ip_cnt;
231 uint64_t psb_ip;
232 const unsigned char *next_buf;
233 size_t next_len;
234 unsigned char temp_buf[INTEL_PT_PKT_MAX_SZ];
235 };
236
intel_pt_lower_power_of_2(uint64_t x)237 static uint64_t intel_pt_lower_power_of_2(uint64_t x)
238 {
239 int i;
240
241 for (i = 0; x != 1; i++)
242 x >>= 1;
243
244 return x << i;
245 }
246
247 __printf(1, 2)
p_log(const char * fmt,...)248 static void p_log(const char *fmt, ...)
249 {
250 char buf[512];
251 va_list args;
252
253 va_start(args, fmt);
254 vsnprintf(buf, sizeof(buf), fmt, args);
255 va_end(args);
256
257 fprintf(stderr, "%s\n", buf);
258 intel_pt_log("%s\n", buf);
259 }
260
intel_pt_print_once(struct intel_pt_decoder * decoder,enum intel_pt_p_once id)261 static bool intel_pt_print_once(struct intel_pt_decoder *decoder,
262 enum intel_pt_p_once id)
263 {
264 uint64_t bit = 1ULL << id;
265
266 if (decoder->print_once & bit)
267 return false;
268 decoder->print_once |= bit;
269 return true;
270 }
271
intel_pt_cyc_threshold(uint64_t ctl)272 static uint64_t intel_pt_cyc_threshold(uint64_t ctl)
273 {
274 if (!(ctl & INTEL_PT_CYC_ENABLE))
275 return 0;
276
277 return (ctl & INTEL_PT_CYC_THRESHOLD) >> INTEL_PT_CYC_THRESHOLD_SHIFT;
278 }
279
intel_pt_setup_period(struct intel_pt_decoder * decoder)280 static void intel_pt_setup_period(struct intel_pt_decoder *decoder)
281 {
282 if (decoder->period_type == INTEL_PT_PERIOD_TICKS) {
283 uint64_t period;
284
285 period = intel_pt_lower_power_of_2(decoder->period);
286 decoder->period_mask = ~(period - 1);
287 decoder->period_ticks = period;
288 }
289 }
290
multdiv(uint64_t t,uint32_t n,uint32_t d)291 static uint64_t multdiv(uint64_t t, uint32_t n, uint32_t d)
292 {
293 if (!d)
294 return 0;
295 return (t / d) * n + ((t % d) * n) / d;
296 }
297
intel_pt_decoder_new(struct intel_pt_params * params)298 struct intel_pt_decoder *intel_pt_decoder_new(struct intel_pt_params *params)
299 {
300 struct intel_pt_decoder *decoder;
301
302 if (!params->get_trace || !params->walk_insn)
303 return NULL;
304
305 decoder = zalloc(sizeof(struct intel_pt_decoder));
306 if (!decoder)
307 return NULL;
308
309 decoder->get_trace = params->get_trace;
310 decoder->walk_insn = params->walk_insn;
311 decoder->pgd_ip = params->pgd_ip;
312 decoder->lookahead = params->lookahead;
313 decoder->findnew_vmcs_info = params->findnew_vmcs_info;
314 decoder->data = params->data;
315 decoder->return_compression = params->return_compression;
316 decoder->branch_enable = params->branch_enable;
317 decoder->hop = params->quick >= 1;
318 decoder->leap = params->quick >= 2;
319 decoder->vm_time_correlation = params->vm_time_correlation;
320 decoder->vm_tm_corr_dry_run = params->vm_tm_corr_dry_run;
321 decoder->first_timestamp = params->first_timestamp;
322 decoder->last_reliable_timestamp = params->first_timestamp;
323 decoder->max_loops = params->max_loops ? params->max_loops : INTEL_PT_MAX_LOOPS;
324
325 decoder->flags = params->flags;
326
327 decoder->ctl = params->ctl;
328 decoder->period = params->period;
329 decoder->period_type = params->period_type;
330
331 decoder->max_non_turbo_ratio = params->max_non_turbo_ratio;
332 decoder->max_non_turbo_ratio_fp = params->max_non_turbo_ratio;
333
334 decoder->cyc_threshold = intel_pt_cyc_threshold(decoder->ctl);
335
336 intel_pt_setup_period(decoder);
337
338 decoder->mtc_shift = params->mtc_period;
339 decoder->ctc_rem_mask = (1 << decoder->mtc_shift) - 1;
340
341 decoder->tsc_ctc_ratio_n = params->tsc_ctc_ratio_n;
342 decoder->tsc_ctc_ratio_d = params->tsc_ctc_ratio_d;
343
344 if (!decoder->tsc_ctc_ratio_n)
345 decoder->tsc_ctc_ratio_d = 0;
346
347 if (decoder->tsc_ctc_ratio_d) {
348 if (!(decoder->tsc_ctc_ratio_n % decoder->tsc_ctc_ratio_d))
349 decoder->tsc_ctc_mult = decoder->tsc_ctc_ratio_n /
350 decoder->tsc_ctc_ratio_d;
351 }
352
353 /*
354 * A TSC packet can slip past MTC packets so that the timestamp appears
355 * to go backwards. One estimate is that can be up to about 40 CPU
356 * cycles, which is certainly less than 0x1000 TSC ticks, but accept
357 * slippage an order of magnitude more to be on the safe side.
358 */
359 decoder->tsc_slip = 0x10000;
360
361 intel_pt_log("timestamp: mtc_shift %u\n", decoder->mtc_shift);
362 intel_pt_log("timestamp: tsc_ctc_ratio_n %u\n", decoder->tsc_ctc_ratio_n);
363 intel_pt_log("timestamp: tsc_ctc_ratio_d %u\n", decoder->tsc_ctc_ratio_d);
364 intel_pt_log("timestamp: tsc_ctc_mult %u\n", decoder->tsc_ctc_mult);
365 intel_pt_log("timestamp: tsc_slip %#x\n", decoder->tsc_slip);
366
367 if (decoder->hop)
368 intel_pt_log("Hop mode: decoding FUP and TIPs, but not TNT\n");
369
370 return decoder;
371 }
372
intel_pt_set_first_timestamp(struct intel_pt_decoder * decoder,uint64_t first_timestamp)373 void intel_pt_set_first_timestamp(struct intel_pt_decoder *decoder,
374 uint64_t first_timestamp)
375 {
376 decoder->first_timestamp = first_timestamp;
377 }
378
intel_pt_pop_blk(struct intel_pt_stack * stack)379 static void intel_pt_pop_blk(struct intel_pt_stack *stack)
380 {
381 struct intel_pt_blk *blk = stack->blk;
382
383 stack->blk = blk->prev;
384 if (!stack->spare)
385 stack->spare = blk;
386 else
387 free(blk);
388 }
389
intel_pt_pop(struct intel_pt_stack * stack)390 static uint64_t intel_pt_pop(struct intel_pt_stack *stack)
391 {
392 if (!stack->pos) {
393 if (!stack->blk)
394 return 0;
395 intel_pt_pop_blk(stack);
396 if (!stack->blk)
397 return 0;
398 stack->pos = INTEL_PT_BLK_SIZE;
399 }
400 return stack->blk->ip[--stack->pos];
401 }
402
intel_pt_alloc_blk(struct intel_pt_stack * stack)403 static int intel_pt_alloc_blk(struct intel_pt_stack *stack)
404 {
405 struct intel_pt_blk *blk;
406
407 if (stack->spare) {
408 blk = stack->spare;
409 stack->spare = NULL;
410 } else {
411 blk = malloc(sizeof(struct intel_pt_blk));
412 if (!blk)
413 return -ENOMEM;
414 }
415
416 blk->prev = stack->blk;
417 stack->blk = blk;
418 stack->pos = 0;
419 return 0;
420 }
421
intel_pt_push(struct intel_pt_stack * stack,uint64_t ip)422 static int intel_pt_push(struct intel_pt_stack *stack, uint64_t ip)
423 {
424 int err;
425
426 if (!stack->blk || stack->pos == INTEL_PT_BLK_SIZE) {
427 err = intel_pt_alloc_blk(stack);
428 if (err)
429 return err;
430 }
431
432 stack->blk->ip[stack->pos++] = ip;
433 return 0;
434 }
435
intel_pt_clear_stack(struct intel_pt_stack * stack)436 static void intel_pt_clear_stack(struct intel_pt_stack *stack)
437 {
438 while (stack->blk)
439 intel_pt_pop_blk(stack);
440 stack->pos = 0;
441 }
442
intel_pt_free_stack(struct intel_pt_stack * stack)443 static void intel_pt_free_stack(struct intel_pt_stack *stack)
444 {
445 intel_pt_clear_stack(stack);
446 zfree(&stack->blk);
447 zfree(&stack->spare);
448 }
449
intel_pt_decoder_free(struct intel_pt_decoder * decoder)450 void intel_pt_decoder_free(struct intel_pt_decoder *decoder)
451 {
452 intel_pt_free_stack(&decoder->stack);
453 free(decoder);
454 }
455
intel_pt_ext_err(int code)456 static int intel_pt_ext_err(int code)
457 {
458 switch (code) {
459 case -ENOMEM:
460 return INTEL_PT_ERR_NOMEM;
461 case -ENOSYS:
462 return INTEL_PT_ERR_INTERN;
463 case -EBADMSG:
464 return INTEL_PT_ERR_BADPKT;
465 case -ENODATA:
466 return INTEL_PT_ERR_NODATA;
467 case -EILSEQ:
468 return INTEL_PT_ERR_NOINSN;
469 case -ENOENT:
470 return INTEL_PT_ERR_MISMAT;
471 case -EOVERFLOW:
472 return INTEL_PT_ERR_OVR;
473 case -ENOSPC:
474 return INTEL_PT_ERR_LOST;
475 case -ELOOP:
476 return INTEL_PT_ERR_NELOOP;
477 case -ECONNRESET:
478 return INTEL_PT_ERR_EPTW;
479 default:
480 return INTEL_PT_ERR_UNK;
481 }
482 }
483
484 static const char *intel_pt_err_msgs[] = {
485 [INTEL_PT_ERR_NOMEM] = "Memory allocation failed",
486 [INTEL_PT_ERR_INTERN] = "Internal error",
487 [INTEL_PT_ERR_BADPKT] = "Bad packet",
488 [INTEL_PT_ERR_NODATA] = "No more data",
489 [INTEL_PT_ERR_NOINSN] = "Failed to get instruction",
490 [INTEL_PT_ERR_MISMAT] = "Trace doesn't match instruction",
491 [INTEL_PT_ERR_OVR] = "Overflow packet",
492 [INTEL_PT_ERR_LOST] = "Lost trace data",
493 [INTEL_PT_ERR_UNK] = "Unknown error!",
494 [INTEL_PT_ERR_NELOOP] = "Never-ending loop (refer perf config intel-pt.max-loops)",
495 [INTEL_PT_ERR_EPTW] = "Broken emulated ptwrite",
496 };
497
intel_pt__strerror(int code,char * buf,size_t buflen)498 int intel_pt__strerror(int code, char *buf, size_t buflen)
499 {
500 if (code < 1 || code >= INTEL_PT_ERR_MAX)
501 code = INTEL_PT_ERR_UNK;
502 strlcpy(buf, intel_pt_err_msgs[code], buflen);
503 return 0;
504 }
505
intel_pt_calc_ip(const struct intel_pt_pkt * packet,uint64_t last_ip)506 static uint64_t intel_pt_calc_ip(const struct intel_pt_pkt *packet,
507 uint64_t last_ip)
508 {
509 uint64_t ip;
510
511 switch (packet->count) {
512 case 1:
513 ip = (last_ip & (uint64_t)0xffffffffffff0000ULL) |
514 packet->payload;
515 break;
516 case 2:
517 ip = (last_ip & (uint64_t)0xffffffff00000000ULL) |
518 packet->payload;
519 break;
520 case 3:
521 ip = packet->payload;
522 /* Sign-extend 6-byte ip */
523 if (ip & (uint64_t)0x800000000000ULL)
524 ip |= (uint64_t)0xffff000000000000ULL;
525 break;
526 case 4:
527 ip = (last_ip & (uint64_t)0xffff000000000000ULL) |
528 packet->payload;
529 break;
530 case 6:
531 ip = packet->payload;
532 break;
533 default:
534 return 0;
535 }
536
537 return ip;
538 }
539
intel_pt_set_last_ip(struct intel_pt_decoder * decoder)540 static inline void intel_pt_set_last_ip(struct intel_pt_decoder *decoder)
541 {
542 decoder->last_ip = intel_pt_calc_ip(&decoder->packet, decoder->last_ip);
543 decoder->have_last_ip = true;
544 }
545
intel_pt_set_ip(struct intel_pt_decoder * decoder)546 static inline void intel_pt_set_ip(struct intel_pt_decoder *decoder)
547 {
548 intel_pt_set_last_ip(decoder);
549 decoder->ip = decoder->last_ip;
550 }
551
intel_pt_decoder_log_packet(struct intel_pt_decoder * decoder)552 static void intel_pt_decoder_log_packet(struct intel_pt_decoder *decoder)
553 {
554 intel_pt_log_packet(&decoder->packet, decoder->pkt_len, decoder->pos,
555 decoder->buf);
556 }
557
intel_pt_bug(struct intel_pt_decoder * decoder)558 static int intel_pt_bug(struct intel_pt_decoder *decoder)
559 {
560 intel_pt_log("ERROR: Internal error\n");
561 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
562 return -ENOSYS;
563 }
564
intel_pt_clear_tx_flags(struct intel_pt_decoder * decoder)565 static inline void intel_pt_clear_tx_flags(struct intel_pt_decoder *decoder)
566 {
567 decoder->tx_flags = 0;
568 }
569
intel_pt_update_in_tx(struct intel_pt_decoder * decoder)570 static inline void intel_pt_update_in_tx(struct intel_pt_decoder *decoder)
571 {
572 decoder->tx_flags = decoder->packet.payload & INTEL_PT_IN_TX;
573 }
574
intel_pt_update_pip(struct intel_pt_decoder * decoder)575 static inline void intel_pt_update_pip(struct intel_pt_decoder *decoder)
576 {
577 decoder->pip_payload = decoder->packet.payload;
578 }
579
intel_pt_update_nr(struct intel_pt_decoder * decoder)580 static inline void intel_pt_update_nr(struct intel_pt_decoder *decoder)
581 {
582 decoder->next_nr = decoder->pip_payload & 1;
583 }
584
intel_pt_set_nr(struct intel_pt_decoder * decoder)585 static inline void intel_pt_set_nr(struct intel_pt_decoder *decoder)
586 {
587 decoder->nr = decoder->pip_payload & 1;
588 decoder->next_nr = decoder->nr;
589 }
590
intel_pt_set_pip(struct intel_pt_decoder * decoder)591 static inline void intel_pt_set_pip(struct intel_pt_decoder *decoder)
592 {
593 intel_pt_update_pip(decoder);
594 intel_pt_set_nr(decoder);
595 }
596
intel_pt_bad_packet(struct intel_pt_decoder * decoder)597 static int intel_pt_bad_packet(struct intel_pt_decoder *decoder)
598 {
599 intel_pt_clear_tx_flags(decoder);
600 decoder->have_tma = false;
601 decoder->pkt_len = 1;
602 decoder->pkt_step = 1;
603 intel_pt_decoder_log_packet(decoder);
604 if (decoder->pkt_state != INTEL_PT_STATE_NO_PSB) {
605 intel_pt_log("ERROR: Bad packet\n");
606 decoder->pkt_state = INTEL_PT_STATE_ERR1;
607 }
608 return -EBADMSG;
609 }
610
intel_pt_update_sample_time(struct intel_pt_decoder * decoder)611 static inline void intel_pt_update_sample_time(struct intel_pt_decoder *decoder)
612 {
613 decoder->sample_timestamp = decoder->timestamp;
614 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
615 }
616
intel_pt_reposition(struct intel_pt_decoder * decoder)617 static void intel_pt_reposition(struct intel_pt_decoder *decoder)
618 {
619 decoder->ip = 0;
620 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
621 decoder->timestamp = 0;
622 decoder->have_tma = false;
623 }
624
intel_pt_get_data(struct intel_pt_decoder * decoder,bool reposition)625 static int intel_pt_get_data(struct intel_pt_decoder *decoder, bool reposition)
626 {
627 struct intel_pt_buffer buffer = { .buf = 0, };
628 int ret;
629
630 decoder->pkt_step = 0;
631
632 intel_pt_log("Getting more data\n");
633 ret = decoder->get_trace(&buffer, decoder->data);
634 if (ret)
635 return ret;
636 decoder->buf = buffer.buf;
637 decoder->len = buffer.len;
638 if (!decoder->len) {
639 intel_pt_log("No more data\n");
640 return -ENODATA;
641 }
642 decoder->buf_timestamp = buffer.ref_timestamp;
643 if (!buffer.consecutive || reposition) {
644 intel_pt_reposition(decoder);
645 decoder->ref_timestamp = buffer.ref_timestamp;
646 decoder->state.trace_nr = buffer.trace_nr;
647 decoder->vm_tm_corr_same_buf = false;
648 intel_pt_log("Reference timestamp 0x%" PRIx64 "\n",
649 decoder->ref_timestamp);
650 return -ENOLINK;
651 }
652
653 return 0;
654 }
655
intel_pt_get_next_data(struct intel_pt_decoder * decoder,bool reposition)656 static int intel_pt_get_next_data(struct intel_pt_decoder *decoder,
657 bool reposition)
658 {
659 if (!decoder->next_buf)
660 return intel_pt_get_data(decoder, reposition);
661
662 decoder->buf = decoder->next_buf;
663 decoder->len = decoder->next_len;
664 decoder->next_buf = 0;
665 decoder->next_len = 0;
666 return 0;
667 }
668
intel_pt_get_split_packet(struct intel_pt_decoder * decoder)669 static int intel_pt_get_split_packet(struct intel_pt_decoder *decoder)
670 {
671 unsigned char *buf = decoder->temp_buf;
672 size_t old_len, len, n;
673 int ret;
674
675 old_len = decoder->len;
676 len = decoder->len;
677 memcpy(buf, decoder->buf, len);
678
679 ret = intel_pt_get_data(decoder, false);
680 if (ret) {
681 decoder->pos += old_len;
682 return ret < 0 ? ret : -EINVAL;
683 }
684
685 n = INTEL_PT_PKT_MAX_SZ - len;
686 if (n > decoder->len)
687 n = decoder->len;
688 memcpy(buf + len, decoder->buf, n);
689 len += n;
690
691 decoder->prev_pkt_ctx = decoder->pkt_ctx;
692 ret = intel_pt_get_packet(buf, len, &decoder->packet, &decoder->pkt_ctx);
693 if (ret < (int)old_len) {
694 decoder->next_buf = decoder->buf;
695 decoder->next_len = decoder->len;
696 decoder->buf = buf;
697 decoder->len = old_len;
698 return intel_pt_bad_packet(decoder);
699 }
700
701 decoder->next_buf = decoder->buf + (ret - old_len);
702 decoder->next_len = decoder->len - (ret - old_len);
703
704 decoder->buf = buf;
705 decoder->len = ret;
706
707 return ret;
708 }
709
710 struct intel_pt_pkt_info {
711 struct intel_pt_decoder *decoder;
712 struct intel_pt_pkt packet;
713 uint64_t pos;
714 int pkt_len;
715 int last_packet_type;
716 void *data;
717 };
718
719 typedef int (*intel_pt_pkt_cb_t)(struct intel_pt_pkt_info *pkt_info);
720
721 /* Lookahead packets in current buffer */
intel_pt_pkt_lookahead(struct intel_pt_decoder * decoder,intel_pt_pkt_cb_t cb,void * data)722 static int intel_pt_pkt_lookahead(struct intel_pt_decoder *decoder,
723 intel_pt_pkt_cb_t cb, void *data)
724 {
725 struct intel_pt_pkt_info pkt_info;
726 const unsigned char *buf = decoder->buf;
727 enum intel_pt_pkt_ctx pkt_ctx = decoder->pkt_ctx;
728 size_t len = decoder->len;
729 int ret;
730
731 pkt_info.decoder = decoder;
732 pkt_info.pos = decoder->pos;
733 pkt_info.pkt_len = decoder->pkt_step;
734 pkt_info.last_packet_type = decoder->last_packet_type;
735 pkt_info.data = data;
736
737 while (1) {
738 do {
739 pkt_info.pos += pkt_info.pkt_len;
740 buf += pkt_info.pkt_len;
741 len -= pkt_info.pkt_len;
742
743 if (!len)
744 return INTEL_PT_NEED_MORE_BYTES;
745
746 ret = intel_pt_get_packet(buf, len, &pkt_info.packet,
747 &pkt_ctx);
748 if (!ret)
749 return INTEL_PT_NEED_MORE_BYTES;
750 if (ret < 0)
751 return ret;
752
753 pkt_info.pkt_len = ret;
754 } while (pkt_info.packet.type == INTEL_PT_PAD);
755
756 ret = cb(&pkt_info);
757 if (ret)
758 return 0;
759
760 pkt_info.last_packet_type = pkt_info.packet.type;
761 }
762 }
763
764 struct intel_pt_calc_cyc_to_tsc_info {
765 uint64_t cycle_cnt;
766 unsigned int cbr;
767 uint32_t last_mtc;
768 uint64_t ctc_timestamp;
769 uint64_t ctc_delta;
770 uint64_t tsc_timestamp;
771 uint64_t timestamp;
772 bool have_tma;
773 bool fixup_last_mtc;
774 bool from_mtc;
775 double cbr_cyc_to_tsc;
776 };
777
778 /*
779 * MTC provides a 8-bit slice of CTC but the TMA packet only provides the lower
780 * 16 bits of CTC. If mtc_shift > 8 then some of the MTC bits are not in the CTC
781 * provided by the TMA packet. Fix-up the last_mtc calculated from the TMA
782 * packet by copying the missing bits from the current MTC assuming the least
783 * difference between the two, and that the current MTC comes after last_mtc.
784 */
intel_pt_fixup_last_mtc(uint32_t mtc,int mtc_shift,uint32_t * last_mtc)785 static void intel_pt_fixup_last_mtc(uint32_t mtc, int mtc_shift,
786 uint32_t *last_mtc)
787 {
788 uint32_t first_missing_bit = 1U << (16 - mtc_shift);
789 uint32_t mask = ~(first_missing_bit - 1);
790
791 *last_mtc |= mtc & mask;
792 if (*last_mtc >= mtc) {
793 *last_mtc -= first_missing_bit;
794 *last_mtc &= 0xff;
795 }
796 }
797
intel_pt_calc_cyc_cb(struct intel_pt_pkt_info * pkt_info)798 static int intel_pt_calc_cyc_cb(struct intel_pt_pkt_info *pkt_info)
799 {
800 struct intel_pt_decoder *decoder = pkt_info->decoder;
801 struct intel_pt_calc_cyc_to_tsc_info *data = pkt_info->data;
802 uint64_t timestamp;
803 double cyc_to_tsc;
804 unsigned int cbr;
805 uint32_t mtc, mtc_delta, ctc, fc, ctc_rem;
806
807 switch (pkt_info->packet.type) {
808 case INTEL_PT_TNT:
809 case INTEL_PT_TIP_PGE:
810 case INTEL_PT_TIP:
811 case INTEL_PT_FUP:
812 case INTEL_PT_PSB:
813 case INTEL_PT_PIP:
814 case INTEL_PT_MODE_EXEC:
815 case INTEL_PT_MODE_TSX:
816 case INTEL_PT_PSBEND:
817 case INTEL_PT_PAD:
818 case INTEL_PT_VMCS:
819 case INTEL_PT_MNT:
820 case INTEL_PT_PTWRITE:
821 case INTEL_PT_PTWRITE_IP:
822 case INTEL_PT_BBP:
823 case INTEL_PT_BIP:
824 case INTEL_PT_BEP:
825 case INTEL_PT_BEP_IP:
826 case INTEL_PT_CFE:
827 case INTEL_PT_CFE_IP:
828 case INTEL_PT_EVD:
829 return 0;
830
831 case INTEL_PT_MTC:
832 if (!data->have_tma)
833 return 0;
834
835 mtc = pkt_info->packet.payload;
836 if (decoder->mtc_shift > 8 && data->fixup_last_mtc) {
837 data->fixup_last_mtc = false;
838 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
839 &data->last_mtc);
840 }
841 if (mtc > data->last_mtc)
842 mtc_delta = mtc - data->last_mtc;
843 else
844 mtc_delta = mtc + 256 - data->last_mtc;
845 data->ctc_delta += mtc_delta << decoder->mtc_shift;
846 data->last_mtc = mtc;
847
848 if (decoder->tsc_ctc_mult) {
849 timestamp = data->ctc_timestamp +
850 data->ctc_delta * decoder->tsc_ctc_mult;
851 } else {
852 timestamp = data->ctc_timestamp +
853 multdiv(data->ctc_delta,
854 decoder->tsc_ctc_ratio_n,
855 decoder->tsc_ctc_ratio_d);
856 }
857
858 if (timestamp < data->timestamp)
859 return 1;
860
861 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
862 data->timestamp = timestamp;
863 return 0;
864 }
865
866 break;
867
868 case INTEL_PT_TSC:
869 /*
870 * For now, do not support using TSC packets - refer
871 * intel_pt_calc_cyc_to_tsc().
872 */
873 if (data->from_mtc)
874 return 1;
875 timestamp = pkt_info->packet.payload |
876 (data->timestamp & (0xffULL << 56));
877 if (data->from_mtc && timestamp < data->timestamp &&
878 data->timestamp - timestamp < decoder->tsc_slip)
879 return 1;
880 if (timestamp < data->timestamp)
881 timestamp += (1ULL << 56);
882 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
883 if (data->from_mtc)
884 return 1;
885 data->tsc_timestamp = timestamp;
886 data->timestamp = timestamp;
887 return 0;
888 }
889 break;
890
891 case INTEL_PT_TMA:
892 if (data->from_mtc)
893 return 1;
894
895 if (!decoder->tsc_ctc_ratio_d)
896 return 0;
897
898 ctc = pkt_info->packet.payload;
899 fc = pkt_info->packet.count;
900 ctc_rem = ctc & decoder->ctc_rem_mask;
901
902 data->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
903
904 data->ctc_timestamp = data->tsc_timestamp - fc;
905 if (decoder->tsc_ctc_mult) {
906 data->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
907 } else {
908 data->ctc_timestamp -=
909 multdiv(ctc_rem, decoder->tsc_ctc_ratio_n,
910 decoder->tsc_ctc_ratio_d);
911 }
912
913 data->ctc_delta = 0;
914 data->have_tma = true;
915 data->fixup_last_mtc = true;
916
917 return 0;
918
919 case INTEL_PT_CYC:
920 data->cycle_cnt += pkt_info->packet.payload;
921 return 0;
922
923 case INTEL_PT_CBR:
924 cbr = pkt_info->packet.payload;
925 if (data->cbr && data->cbr != cbr)
926 return 1;
927 data->cbr = cbr;
928 data->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
929 return 0;
930
931 case INTEL_PT_TIP_PGD:
932 case INTEL_PT_TRACESTOP:
933 case INTEL_PT_EXSTOP:
934 case INTEL_PT_EXSTOP_IP:
935 case INTEL_PT_MWAIT:
936 case INTEL_PT_PWRE:
937 case INTEL_PT_PWRX:
938 case INTEL_PT_OVF:
939 case INTEL_PT_BAD: /* Does not happen */
940 default:
941 return 1;
942 }
943
944 if (!data->cbr && decoder->cbr) {
945 data->cbr = decoder->cbr;
946 data->cbr_cyc_to_tsc = decoder->cbr_cyc_to_tsc;
947 }
948
949 if (!data->cycle_cnt)
950 return 1;
951
952 cyc_to_tsc = (double)(timestamp - decoder->timestamp) / data->cycle_cnt;
953
954 if (data->cbr && cyc_to_tsc > data->cbr_cyc_to_tsc &&
955 cyc_to_tsc / data->cbr_cyc_to_tsc > 1.25) {
956 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle too big (c.f. CBR-based value %g), pos " x64_fmt "\n",
957 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
958 return 1;
959 }
960
961 decoder->calc_cyc_to_tsc = cyc_to_tsc;
962 decoder->have_calc_cyc_to_tsc = true;
963
964 if (data->cbr) {
965 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. CBR-based value %g, pos " x64_fmt "\n",
966 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
967 } else {
968 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. unknown CBR-based value, pos " x64_fmt "\n",
969 cyc_to_tsc, pkt_info->pos);
970 }
971
972 return 1;
973 }
974
intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder * decoder,bool from_mtc)975 static void intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder *decoder,
976 bool from_mtc)
977 {
978 struct intel_pt_calc_cyc_to_tsc_info data = {
979 .cycle_cnt = 0,
980 .cbr = 0,
981 .last_mtc = decoder->last_mtc,
982 .ctc_timestamp = decoder->ctc_timestamp,
983 .ctc_delta = decoder->ctc_delta,
984 .tsc_timestamp = decoder->tsc_timestamp,
985 .timestamp = decoder->timestamp,
986 .have_tma = decoder->have_tma,
987 .fixup_last_mtc = decoder->fixup_last_mtc,
988 .from_mtc = from_mtc,
989 .cbr_cyc_to_tsc = 0,
990 };
991
992 /*
993 * For now, do not support using TSC packets for at least the reasons:
994 * 1) timing might have stopped
995 * 2) TSC packets within PSB+ can slip against CYC packets
996 */
997 if (!from_mtc)
998 return;
999
1000 intel_pt_pkt_lookahead(decoder, intel_pt_calc_cyc_cb, &data);
1001 }
1002
intel_pt_get_next_packet(struct intel_pt_decoder * decoder)1003 static int intel_pt_get_next_packet(struct intel_pt_decoder *decoder)
1004 {
1005 int ret;
1006
1007 decoder->last_packet_type = decoder->packet.type;
1008
1009 do {
1010 decoder->pos += decoder->pkt_step;
1011 decoder->buf += decoder->pkt_step;
1012 decoder->len -= decoder->pkt_step;
1013
1014 if (!decoder->len) {
1015 ret = intel_pt_get_next_data(decoder, false);
1016 if (ret)
1017 return ret;
1018 }
1019
1020 decoder->prev_pkt_ctx = decoder->pkt_ctx;
1021 ret = intel_pt_get_packet(decoder->buf, decoder->len,
1022 &decoder->packet, &decoder->pkt_ctx);
1023 if (ret == INTEL_PT_NEED_MORE_BYTES && BITS_PER_LONG == 32 &&
1024 decoder->len < INTEL_PT_PKT_MAX_SZ && !decoder->next_buf) {
1025 ret = intel_pt_get_split_packet(decoder);
1026 if (ret < 0)
1027 return ret;
1028 }
1029 if (ret <= 0)
1030 return intel_pt_bad_packet(decoder);
1031
1032 decoder->pkt_len = ret;
1033 decoder->pkt_step = ret;
1034 intel_pt_decoder_log_packet(decoder);
1035 } while (decoder->packet.type == INTEL_PT_PAD);
1036
1037 return 0;
1038 }
1039
intel_pt_next_period(struct intel_pt_decoder * decoder)1040 static uint64_t intel_pt_next_period(struct intel_pt_decoder *decoder)
1041 {
1042 uint64_t timestamp, masked_timestamp;
1043
1044 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
1045 masked_timestamp = timestamp & decoder->period_mask;
1046 if (decoder->continuous_period) {
1047 if (masked_timestamp > decoder->last_masked_timestamp)
1048 return 1;
1049 } else {
1050 timestamp += 1;
1051 masked_timestamp = timestamp & decoder->period_mask;
1052 if (masked_timestamp > decoder->last_masked_timestamp) {
1053 decoder->last_masked_timestamp = masked_timestamp;
1054 decoder->continuous_period = true;
1055 }
1056 }
1057
1058 if (masked_timestamp < decoder->last_masked_timestamp)
1059 return decoder->period_ticks;
1060
1061 return decoder->period_ticks - (timestamp - masked_timestamp);
1062 }
1063
intel_pt_next_sample(struct intel_pt_decoder * decoder)1064 static uint64_t intel_pt_next_sample(struct intel_pt_decoder *decoder)
1065 {
1066 switch (decoder->period_type) {
1067 case INTEL_PT_PERIOD_INSTRUCTIONS:
1068 return decoder->period - decoder->period_insn_cnt;
1069 case INTEL_PT_PERIOD_TICKS:
1070 return intel_pt_next_period(decoder);
1071 case INTEL_PT_PERIOD_NONE:
1072 case INTEL_PT_PERIOD_MTC:
1073 default:
1074 return 0;
1075 }
1076 }
1077
intel_pt_sample_insn(struct intel_pt_decoder * decoder)1078 static void intel_pt_sample_insn(struct intel_pt_decoder *decoder)
1079 {
1080 uint64_t timestamp, masked_timestamp;
1081
1082 switch (decoder->period_type) {
1083 case INTEL_PT_PERIOD_INSTRUCTIONS:
1084 decoder->period_insn_cnt = 0;
1085 break;
1086 case INTEL_PT_PERIOD_TICKS:
1087 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
1088 masked_timestamp = timestamp & decoder->period_mask;
1089 if (masked_timestamp > decoder->last_masked_timestamp)
1090 decoder->last_masked_timestamp = masked_timestamp;
1091 else
1092 decoder->last_masked_timestamp += decoder->period_ticks;
1093 break;
1094 case INTEL_PT_PERIOD_NONE:
1095 case INTEL_PT_PERIOD_MTC:
1096 default:
1097 break;
1098 }
1099
1100 decoder->state.type |= INTEL_PT_INSTRUCTION;
1101 }
1102
intel_pt_walk_insn(struct intel_pt_decoder * decoder,struct intel_pt_insn * intel_pt_insn,uint64_t ip)1103 static int intel_pt_walk_insn(struct intel_pt_decoder *decoder,
1104 struct intel_pt_insn *intel_pt_insn, uint64_t ip)
1105 {
1106 uint64_t max_insn_cnt, insn_cnt = 0;
1107 int err;
1108
1109 if (!decoder->mtc_insn)
1110 decoder->mtc_insn = true;
1111
1112 max_insn_cnt = intel_pt_next_sample(decoder);
1113
1114 err = decoder->walk_insn(intel_pt_insn, &insn_cnt, &decoder->ip, ip,
1115 max_insn_cnt, decoder->data);
1116
1117 decoder->tot_insn_cnt += insn_cnt;
1118 decoder->timestamp_insn_cnt += insn_cnt;
1119 decoder->sample_insn_cnt += insn_cnt;
1120 decoder->period_insn_cnt += insn_cnt;
1121
1122 if (err) {
1123 decoder->no_progress = 0;
1124 decoder->pkt_state = INTEL_PT_STATE_ERR2;
1125 intel_pt_log_at("ERROR: Failed to get instruction",
1126 decoder->ip);
1127 if (err == -ENOENT)
1128 return -ENOLINK;
1129 return -EILSEQ;
1130 }
1131
1132 if (ip && decoder->ip == ip) {
1133 err = -EAGAIN;
1134 goto out;
1135 }
1136
1137 if (max_insn_cnt && insn_cnt >= max_insn_cnt)
1138 intel_pt_sample_insn(decoder);
1139
1140 if (intel_pt_insn->branch == INTEL_PT_BR_NO_BRANCH) {
1141 decoder->state.type = INTEL_PT_INSTRUCTION;
1142 decoder->state.from_ip = decoder->ip;
1143 decoder->state.to_ip = 0;
1144 decoder->ip += intel_pt_insn->length;
1145 err = INTEL_PT_RETURN;
1146 goto out;
1147 }
1148
1149 if (intel_pt_insn->op == INTEL_PT_OP_CALL) {
1150 /* Zero-length calls are excluded */
1151 if (intel_pt_insn->branch != INTEL_PT_BR_UNCONDITIONAL ||
1152 intel_pt_insn->rel) {
1153 err = intel_pt_push(&decoder->stack, decoder->ip +
1154 intel_pt_insn->length);
1155 if (err)
1156 goto out;
1157 }
1158 } else if (intel_pt_insn->op == INTEL_PT_OP_RET) {
1159 decoder->ret_addr = intel_pt_pop(&decoder->stack);
1160 }
1161
1162 if (intel_pt_insn->branch == INTEL_PT_BR_UNCONDITIONAL) {
1163 int cnt = decoder->no_progress++;
1164
1165 decoder->state.from_ip = decoder->ip;
1166 decoder->ip += intel_pt_insn->length +
1167 intel_pt_insn->rel;
1168 decoder->state.to_ip = decoder->ip;
1169 err = INTEL_PT_RETURN;
1170
1171 /*
1172 * Check for being stuck in a loop. This can happen if a
1173 * decoder error results in the decoder erroneously setting the
1174 * ip to an address that is itself in an infinite loop that
1175 * consumes no packets. When that happens, there must be an
1176 * unconditional branch.
1177 */
1178 if (cnt) {
1179 if (cnt == 1) {
1180 decoder->stuck_ip = decoder->state.to_ip;
1181 decoder->stuck_ip_prd = 1;
1182 decoder->stuck_ip_cnt = 1;
1183 } else if (cnt > decoder->max_loops ||
1184 decoder->state.to_ip == decoder->stuck_ip) {
1185 intel_pt_log_at("ERROR: Never-ending loop",
1186 decoder->state.to_ip);
1187 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1188 err = -ELOOP;
1189 goto out;
1190 } else if (!--decoder->stuck_ip_cnt) {
1191 decoder->stuck_ip_prd += 1;
1192 decoder->stuck_ip_cnt = decoder->stuck_ip_prd;
1193 decoder->stuck_ip = decoder->state.to_ip;
1194 }
1195 }
1196 goto out_no_progress;
1197 }
1198 out:
1199 decoder->no_progress = 0;
1200 out_no_progress:
1201 decoder->state.insn_op = intel_pt_insn->op;
1202 decoder->state.insn_len = intel_pt_insn->length;
1203 memcpy(decoder->state.insn, intel_pt_insn->buf,
1204 INTEL_PT_INSN_BUF_SZ);
1205
1206 if (decoder->tx_flags & INTEL_PT_IN_TX)
1207 decoder->state.flags |= INTEL_PT_IN_TX;
1208
1209 return err;
1210 }
1211
intel_pt_fup_event(struct intel_pt_decoder * decoder)1212 static bool intel_pt_fup_event(struct intel_pt_decoder *decoder)
1213 {
1214 enum intel_pt_sample_type type = decoder->state.type;
1215 bool ret = false;
1216
1217 decoder->state.type &= ~INTEL_PT_BRANCH;
1218
1219 if (decoder->set_fup_tx_flags) {
1220 decoder->set_fup_tx_flags = false;
1221 decoder->tx_flags = decoder->fup_tx_flags;
1222 decoder->state.type |= INTEL_PT_TRANSACTION;
1223 if (decoder->fup_tx_flags & INTEL_PT_ABORT_TX)
1224 decoder->state.type |= INTEL_PT_BRANCH;
1225 decoder->state.flags = decoder->fup_tx_flags;
1226 ret = true;
1227 }
1228 if (decoder->set_fup_ptw) {
1229 decoder->set_fup_ptw = false;
1230 decoder->state.type |= INTEL_PT_PTW;
1231 decoder->state.flags |= INTEL_PT_FUP_IP;
1232 decoder->state.ptw_payload = decoder->fup_ptw_payload;
1233 ret = true;
1234 }
1235 if (decoder->set_fup_mwait) {
1236 decoder->set_fup_mwait = false;
1237 decoder->state.type |= INTEL_PT_MWAIT_OP;
1238 decoder->state.mwait_payload = decoder->fup_mwait_payload;
1239 ret = true;
1240 }
1241 if (decoder->set_fup_pwre) {
1242 decoder->set_fup_pwre = false;
1243 decoder->state.type |= INTEL_PT_PWR_ENTRY;
1244 decoder->state.pwre_payload = decoder->fup_pwre_payload;
1245 ret = true;
1246 }
1247 if (decoder->set_fup_exstop) {
1248 decoder->set_fup_exstop = false;
1249 decoder->state.type |= INTEL_PT_EX_STOP;
1250 decoder->state.flags |= INTEL_PT_FUP_IP;
1251 ret = true;
1252 }
1253 if (decoder->set_fup_bep) {
1254 decoder->set_fup_bep = false;
1255 decoder->state.type |= INTEL_PT_BLK_ITEMS;
1256 ret = true;
1257 }
1258 if (decoder->overflow) {
1259 decoder->overflow = false;
1260 if (!ret && !decoder->pge) {
1261 if (decoder->hop) {
1262 decoder->state.type = 0;
1263 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
1264 }
1265 decoder->pge = true;
1266 decoder->state.type |= INTEL_PT_BRANCH | INTEL_PT_TRACE_BEGIN;
1267 decoder->state.from_ip = 0;
1268 decoder->state.to_ip = decoder->ip;
1269 return true;
1270 }
1271 }
1272 if (ret) {
1273 decoder->state.from_ip = decoder->ip;
1274 decoder->state.to_ip = 0;
1275 } else {
1276 decoder->state.type = type;
1277 }
1278 return ret;
1279 }
1280
intel_pt_fup_with_nlip(struct intel_pt_decoder * decoder,struct intel_pt_insn * intel_pt_insn,uint64_t ip,int err)1281 static inline bool intel_pt_fup_with_nlip(struct intel_pt_decoder *decoder,
1282 struct intel_pt_insn *intel_pt_insn,
1283 uint64_t ip, int err)
1284 {
1285 return decoder->flags & INTEL_PT_FUP_WITH_NLIP && !err &&
1286 intel_pt_insn->branch == INTEL_PT_BR_INDIRECT &&
1287 ip == decoder->ip + intel_pt_insn->length;
1288 }
1289
intel_pt_walk_fup(struct intel_pt_decoder * decoder)1290 static int intel_pt_walk_fup(struct intel_pt_decoder *decoder)
1291 {
1292 struct intel_pt_insn intel_pt_insn;
1293 uint64_t ip;
1294 int err;
1295
1296 ip = decoder->last_ip;
1297
1298 while (1) {
1299 err = intel_pt_walk_insn(decoder, &intel_pt_insn, ip);
1300 if (err == INTEL_PT_RETURN)
1301 return 0;
1302 if (err == -EAGAIN ||
1303 intel_pt_fup_with_nlip(decoder, &intel_pt_insn, ip, err)) {
1304 bool no_tip = decoder->pkt_state != INTEL_PT_STATE_FUP;
1305
1306 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1307 if (intel_pt_fup_event(decoder) && no_tip)
1308 return 0;
1309 return -EAGAIN;
1310 }
1311 decoder->set_fup_tx_flags = false;
1312 if (err)
1313 return err;
1314
1315 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1316 intel_pt_log_at("ERROR: Unexpected indirect branch",
1317 decoder->ip);
1318 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1319 return -ENOENT;
1320 }
1321
1322 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1323 intel_pt_log_at("ERROR: Unexpected conditional branch",
1324 decoder->ip);
1325 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1326 return -ENOENT;
1327 }
1328
1329 intel_pt_bug(decoder);
1330 }
1331 }
1332
intel_pt_walk_tip(struct intel_pt_decoder * decoder)1333 static int intel_pt_walk_tip(struct intel_pt_decoder *decoder)
1334 {
1335 struct intel_pt_insn intel_pt_insn;
1336 int err;
1337
1338 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1339 if (err == INTEL_PT_RETURN &&
1340 decoder->pgd_ip &&
1341 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1342 (decoder->state.type & INTEL_PT_BRANCH) &&
1343 decoder->pgd_ip(decoder->state.to_ip, decoder->data)) {
1344 /* Unconditional branch leaving filter region */
1345 decoder->no_progress = 0;
1346 decoder->pge = false;
1347 decoder->continuous_period = false;
1348 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1349 decoder->state.type |= INTEL_PT_TRACE_END;
1350 intel_pt_update_nr(decoder);
1351 return 0;
1352 }
1353 if (err == INTEL_PT_RETURN)
1354 return 0;
1355 if (err)
1356 return err;
1357
1358 intel_pt_update_nr(decoder);
1359
1360 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1361 if (decoder->pkt_state == INTEL_PT_STATE_TIP_PGD) {
1362 decoder->pge = false;
1363 decoder->continuous_period = false;
1364 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1365 decoder->state.from_ip = decoder->ip;
1366 if (decoder->packet.count == 0) {
1367 decoder->state.to_ip = 0;
1368 } else {
1369 decoder->state.to_ip = decoder->last_ip;
1370 decoder->ip = decoder->last_ip;
1371 }
1372 decoder->state.type |= INTEL_PT_TRACE_END;
1373 } else {
1374 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1375 decoder->state.from_ip = decoder->ip;
1376 if (decoder->packet.count == 0) {
1377 decoder->state.to_ip = 0;
1378 } else {
1379 decoder->state.to_ip = decoder->last_ip;
1380 decoder->ip = decoder->last_ip;
1381 }
1382 }
1383 return 0;
1384 }
1385
1386 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1387 uint64_t to_ip = decoder->ip + intel_pt_insn.length +
1388 intel_pt_insn.rel;
1389
1390 if (decoder->pgd_ip &&
1391 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1392 decoder->pgd_ip(to_ip, decoder->data)) {
1393 /* Conditional branch leaving filter region */
1394 decoder->pge = false;
1395 decoder->continuous_period = false;
1396 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1397 decoder->ip = to_ip;
1398 decoder->state.from_ip = decoder->ip;
1399 decoder->state.to_ip = to_ip;
1400 decoder->state.type |= INTEL_PT_TRACE_END;
1401 return 0;
1402 }
1403 intel_pt_log_at("ERROR: Conditional branch when expecting indirect branch",
1404 decoder->ip);
1405 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1406 return -ENOENT;
1407 }
1408
1409 return intel_pt_bug(decoder);
1410 }
1411
1412 struct eptw_data {
1413 int bit_countdown;
1414 uint64_t payload;
1415 };
1416
intel_pt_eptw_lookahead_cb(struct intel_pt_pkt_info * pkt_info)1417 static int intel_pt_eptw_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
1418 {
1419 struct eptw_data *data = pkt_info->data;
1420 int nr_bits;
1421
1422 switch (pkt_info->packet.type) {
1423 case INTEL_PT_PAD:
1424 case INTEL_PT_MNT:
1425 case INTEL_PT_MODE_EXEC:
1426 case INTEL_PT_MODE_TSX:
1427 case INTEL_PT_MTC:
1428 case INTEL_PT_FUP:
1429 case INTEL_PT_CYC:
1430 case INTEL_PT_CBR:
1431 case INTEL_PT_TSC:
1432 case INTEL_PT_TMA:
1433 case INTEL_PT_PIP:
1434 case INTEL_PT_VMCS:
1435 case INTEL_PT_PSB:
1436 case INTEL_PT_PSBEND:
1437 case INTEL_PT_PTWRITE:
1438 case INTEL_PT_PTWRITE_IP:
1439 case INTEL_PT_EXSTOP:
1440 case INTEL_PT_EXSTOP_IP:
1441 case INTEL_PT_MWAIT:
1442 case INTEL_PT_PWRE:
1443 case INTEL_PT_PWRX:
1444 case INTEL_PT_BBP:
1445 case INTEL_PT_BIP:
1446 case INTEL_PT_BEP:
1447 case INTEL_PT_BEP_IP:
1448 case INTEL_PT_CFE:
1449 case INTEL_PT_CFE_IP:
1450 case INTEL_PT_EVD:
1451 break;
1452
1453 case INTEL_PT_TNT:
1454 nr_bits = data->bit_countdown;
1455 if (nr_bits > pkt_info->packet.count)
1456 nr_bits = pkt_info->packet.count;
1457 data->payload <<= nr_bits;
1458 data->payload |= pkt_info->packet.payload >> (64 - nr_bits);
1459 data->bit_countdown -= nr_bits;
1460 return !data->bit_countdown;
1461
1462 case INTEL_PT_TIP_PGE:
1463 case INTEL_PT_TIP_PGD:
1464 case INTEL_PT_TIP:
1465 case INTEL_PT_BAD:
1466 case INTEL_PT_OVF:
1467 case INTEL_PT_TRACESTOP:
1468 default:
1469 return 1;
1470 }
1471
1472 return 0;
1473 }
1474
intel_pt_emulated_ptwrite(struct intel_pt_decoder * decoder)1475 static int intel_pt_emulated_ptwrite(struct intel_pt_decoder *decoder)
1476 {
1477 int n = 64 - decoder->tnt.count;
1478 struct eptw_data data = {
1479 .bit_countdown = n,
1480 .payload = decoder->tnt.payload >> n,
1481 };
1482
1483 decoder->emulated_ptwrite = false;
1484 intel_pt_log("Emulated ptwrite detected\n");
1485
1486 intel_pt_pkt_lookahead(decoder, intel_pt_eptw_lookahead_cb, &data);
1487 if (data.bit_countdown)
1488 return -ECONNRESET;
1489
1490 decoder->state.type = INTEL_PT_PTW;
1491 decoder->state.from_ip = decoder->ip;
1492 decoder->state.to_ip = 0;
1493 decoder->state.ptw_payload = data.payload;
1494 return 0;
1495 }
1496
intel_pt_walk_tnt(struct intel_pt_decoder * decoder)1497 static int intel_pt_walk_tnt(struct intel_pt_decoder *decoder)
1498 {
1499 struct intel_pt_insn intel_pt_insn;
1500 int err;
1501
1502 while (1) {
1503 if (decoder->emulated_ptwrite)
1504 return intel_pt_emulated_ptwrite(decoder);
1505 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1506 if (err == INTEL_PT_RETURN) {
1507 decoder->emulated_ptwrite = intel_pt_insn.emulated_ptwrite;
1508 return 0;
1509 }
1510 if (err) {
1511 decoder->emulated_ptwrite = false;
1512 return err;
1513 }
1514
1515 if (intel_pt_insn.op == INTEL_PT_OP_RET) {
1516 if (!decoder->return_compression) {
1517 intel_pt_log_at("ERROR: RET when expecting conditional branch",
1518 decoder->ip);
1519 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1520 return -ENOENT;
1521 }
1522 if (!decoder->ret_addr) {
1523 intel_pt_log_at("ERROR: Bad RET compression (stack empty)",
1524 decoder->ip);
1525 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1526 return -ENOENT;
1527 }
1528 if (!(decoder->tnt.payload & BIT63)) {
1529 intel_pt_log_at("ERROR: Bad RET compression (TNT=N)",
1530 decoder->ip);
1531 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1532 return -ENOENT;
1533 }
1534 decoder->tnt.count -= 1;
1535 if (decoder->tnt.count)
1536 decoder->pkt_state = INTEL_PT_STATE_TNT_CONT;
1537 else
1538 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1539 decoder->tnt.payload <<= 1;
1540 decoder->state.from_ip = decoder->ip;
1541 decoder->ip = decoder->ret_addr;
1542 decoder->state.to_ip = decoder->ip;
1543 return 0;
1544 }
1545
1546 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1547 /* Handle deferred TIPs */
1548 err = intel_pt_get_next_packet(decoder);
1549 if (err)
1550 return err;
1551 if (decoder->packet.type != INTEL_PT_TIP ||
1552 decoder->packet.count == 0) {
1553 intel_pt_log_at("ERROR: Missing deferred TIP for indirect branch",
1554 decoder->ip);
1555 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1556 decoder->pkt_step = 0;
1557 return -ENOENT;
1558 }
1559 intel_pt_set_last_ip(decoder);
1560 decoder->state.from_ip = decoder->ip;
1561 decoder->state.to_ip = decoder->last_ip;
1562 decoder->ip = decoder->last_ip;
1563 intel_pt_update_nr(decoder);
1564 return 0;
1565 }
1566
1567 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1568 decoder->tnt.count -= 1;
1569 if (decoder->tnt.count)
1570 decoder->pkt_state = INTEL_PT_STATE_TNT_CONT;
1571 else
1572 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1573 if (decoder->tnt.payload & BIT63) {
1574 decoder->tnt.payload <<= 1;
1575 decoder->state.from_ip = decoder->ip;
1576 decoder->ip += intel_pt_insn.length +
1577 intel_pt_insn.rel;
1578 decoder->state.to_ip = decoder->ip;
1579 return 0;
1580 }
1581 /* Instruction sample for a non-taken branch */
1582 if (decoder->state.type & INTEL_PT_INSTRUCTION) {
1583 decoder->tnt.payload <<= 1;
1584 decoder->state.type = INTEL_PT_INSTRUCTION;
1585 decoder->state.from_ip = decoder->ip;
1586 decoder->state.to_ip = 0;
1587 decoder->ip += intel_pt_insn.length;
1588 return 0;
1589 }
1590 decoder->sample_cyc = false;
1591 decoder->ip += intel_pt_insn.length;
1592 if (!decoder->tnt.count) {
1593 intel_pt_update_sample_time(decoder);
1594 return -EAGAIN;
1595 }
1596 decoder->tnt.payload <<= 1;
1597 continue;
1598 }
1599
1600 return intel_pt_bug(decoder);
1601 }
1602 }
1603
intel_pt_mode_tsx(struct intel_pt_decoder * decoder,bool * no_tip)1604 static int intel_pt_mode_tsx(struct intel_pt_decoder *decoder, bool *no_tip)
1605 {
1606 unsigned int fup_tx_flags;
1607 int err;
1608
1609 fup_tx_flags = decoder->packet.payload &
1610 (INTEL_PT_IN_TX | INTEL_PT_ABORT_TX);
1611 err = intel_pt_get_next_packet(decoder);
1612 if (err)
1613 return err;
1614 if (decoder->packet.type == INTEL_PT_FUP) {
1615 decoder->fup_tx_flags = fup_tx_flags;
1616 decoder->set_fup_tx_flags = true;
1617 if (!(decoder->fup_tx_flags & INTEL_PT_ABORT_TX))
1618 *no_tip = true;
1619 } else {
1620 intel_pt_log_at("ERROR: Missing FUP after MODE.TSX",
1621 decoder->pos);
1622 intel_pt_update_in_tx(decoder);
1623 }
1624 return 0;
1625 }
1626
intel_pt_8b_tsc(uint64_t timestamp,uint64_t ref_timestamp)1627 static uint64_t intel_pt_8b_tsc(uint64_t timestamp, uint64_t ref_timestamp)
1628 {
1629 timestamp |= (ref_timestamp & (0xffULL << 56));
1630
1631 if (timestamp < ref_timestamp) {
1632 if (ref_timestamp - timestamp > (1ULL << 55))
1633 timestamp += (1ULL << 56);
1634 } else {
1635 if (timestamp - ref_timestamp > (1ULL << 55))
1636 timestamp -= (1ULL << 56);
1637 }
1638
1639 return timestamp;
1640 }
1641
1642 /* For use only when decoder->vm_time_correlation is true */
intel_pt_time_in_range(struct intel_pt_decoder * decoder,uint64_t timestamp)1643 static bool intel_pt_time_in_range(struct intel_pt_decoder *decoder,
1644 uint64_t timestamp)
1645 {
1646 uint64_t max_timestamp = decoder->buf_timestamp;
1647
1648 if (!max_timestamp) {
1649 max_timestamp = decoder->last_reliable_timestamp +
1650 0x400000000ULL;
1651 }
1652 return timestamp >= decoder->last_reliable_timestamp &&
1653 timestamp < decoder->buf_timestamp;
1654 }
1655
intel_pt_calc_tsc_timestamp(struct intel_pt_decoder * decoder)1656 static void intel_pt_calc_tsc_timestamp(struct intel_pt_decoder *decoder)
1657 {
1658 uint64_t timestamp;
1659 bool bad = false;
1660
1661 decoder->have_tma = false;
1662
1663 if (decoder->ref_timestamp) {
1664 timestamp = intel_pt_8b_tsc(decoder->packet.payload,
1665 decoder->ref_timestamp);
1666 decoder->tsc_timestamp = timestamp;
1667 decoder->timestamp = timestamp;
1668 decoder->ref_timestamp = 0;
1669 decoder->timestamp_insn_cnt = 0;
1670 } else if (decoder->timestamp) {
1671 timestamp = decoder->packet.payload |
1672 (decoder->timestamp & (0xffULL << 56));
1673 decoder->tsc_timestamp = timestamp;
1674 if (timestamp < decoder->timestamp &&
1675 decoder->timestamp - timestamp < decoder->tsc_slip) {
1676 intel_pt_log_to("Suppressing backwards timestamp",
1677 timestamp);
1678 timestamp = decoder->timestamp;
1679 }
1680 if (timestamp < decoder->timestamp) {
1681 if (!decoder->buf_timestamp ||
1682 (timestamp + (1ULL << 56) < decoder->buf_timestamp)) {
1683 intel_pt_log_to("Wraparound timestamp", timestamp);
1684 timestamp += (1ULL << 56);
1685 decoder->tsc_timestamp = timestamp;
1686 } else {
1687 intel_pt_log_to("Suppressing bad timestamp", timestamp);
1688 timestamp = decoder->timestamp;
1689 bad = true;
1690 }
1691 }
1692 if (decoder->vm_time_correlation &&
1693 (bad || !intel_pt_time_in_range(decoder, timestamp)) &&
1694 intel_pt_print_once(decoder, INTEL_PT_PRT_ONCE_ERANGE))
1695 p_log("Timestamp out of range");
1696 decoder->timestamp = timestamp;
1697 decoder->timestamp_insn_cnt = 0;
1698 }
1699
1700 if (decoder->last_packet_type == INTEL_PT_CYC) {
1701 decoder->cyc_ref_timestamp = decoder->timestamp;
1702 decoder->cycle_cnt = 0;
1703 decoder->have_calc_cyc_to_tsc = false;
1704 intel_pt_calc_cyc_to_tsc(decoder, false);
1705 }
1706
1707 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1708 }
1709
intel_pt_overflow(struct intel_pt_decoder * decoder)1710 static int intel_pt_overflow(struct intel_pt_decoder *decoder)
1711 {
1712 intel_pt_log("ERROR: Buffer overflow\n");
1713 intel_pt_clear_tx_flags(decoder);
1714 intel_pt_set_nr(decoder);
1715 decoder->timestamp_insn_cnt = 0;
1716 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1717 decoder->state.from_ip = decoder->ip;
1718 decoder->ip = 0;
1719 decoder->pge = false;
1720 decoder->set_fup_tx_flags = false;
1721 decoder->set_fup_ptw = false;
1722 decoder->set_fup_mwait = false;
1723 decoder->set_fup_pwre = false;
1724 decoder->set_fup_exstop = false;
1725 decoder->set_fup_bep = false;
1726 decoder->overflow = true;
1727 return -EOVERFLOW;
1728 }
1729
intel_pt_mtc_cyc_cnt_pge(struct intel_pt_decoder * decoder)1730 static inline void intel_pt_mtc_cyc_cnt_pge(struct intel_pt_decoder *decoder)
1731 {
1732 if (decoder->have_cyc)
1733 return;
1734
1735 decoder->cyc_cnt_timestamp = decoder->timestamp;
1736 decoder->base_cyc_cnt = decoder->tot_cyc_cnt;
1737 }
1738
intel_pt_mtc_cyc_cnt_cbr(struct intel_pt_decoder * decoder)1739 static inline void intel_pt_mtc_cyc_cnt_cbr(struct intel_pt_decoder *decoder)
1740 {
1741 decoder->tsc_to_cyc = decoder->cbr / decoder->max_non_turbo_ratio_fp;
1742
1743 if (decoder->pge)
1744 intel_pt_mtc_cyc_cnt_pge(decoder);
1745 }
1746
intel_pt_mtc_cyc_cnt_upd(struct intel_pt_decoder * decoder)1747 static inline void intel_pt_mtc_cyc_cnt_upd(struct intel_pt_decoder *decoder)
1748 {
1749 uint64_t tot_cyc_cnt, tsc_delta;
1750
1751 if (decoder->have_cyc)
1752 return;
1753
1754 decoder->sample_cyc = true;
1755
1756 if (!decoder->pge || decoder->timestamp <= decoder->cyc_cnt_timestamp)
1757 return;
1758
1759 tsc_delta = decoder->timestamp - decoder->cyc_cnt_timestamp;
1760 tot_cyc_cnt = tsc_delta * decoder->tsc_to_cyc + decoder->base_cyc_cnt;
1761
1762 if (tot_cyc_cnt > decoder->tot_cyc_cnt)
1763 decoder->tot_cyc_cnt = tot_cyc_cnt;
1764 }
1765
intel_pt_calc_tma(struct intel_pt_decoder * decoder)1766 static void intel_pt_calc_tma(struct intel_pt_decoder *decoder)
1767 {
1768 uint32_t ctc = decoder->packet.payload;
1769 uint32_t fc = decoder->packet.count;
1770 uint32_t ctc_rem = ctc & decoder->ctc_rem_mask;
1771
1772 if (!decoder->tsc_ctc_ratio_d)
1773 return;
1774
1775 if (decoder->pge && !decoder->in_psb)
1776 intel_pt_mtc_cyc_cnt_pge(decoder);
1777 else
1778 intel_pt_mtc_cyc_cnt_upd(decoder);
1779
1780 decoder->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
1781 decoder->last_ctc = ctc - ctc_rem;
1782 decoder->ctc_timestamp = decoder->tsc_timestamp - fc;
1783 if (decoder->tsc_ctc_mult) {
1784 decoder->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
1785 } else {
1786 decoder->ctc_timestamp -= multdiv(ctc_rem,
1787 decoder->tsc_ctc_ratio_n,
1788 decoder->tsc_ctc_ratio_d);
1789 }
1790 decoder->ctc_delta = 0;
1791 decoder->have_tma = true;
1792 decoder->fixup_last_mtc = true;
1793 intel_pt_log("CTC timestamp " x64_fmt " last MTC %#x CTC rem %#x\n",
1794 decoder->ctc_timestamp, decoder->last_mtc, ctc_rem);
1795 }
1796
intel_pt_calc_mtc_timestamp(struct intel_pt_decoder * decoder)1797 static void intel_pt_calc_mtc_timestamp(struct intel_pt_decoder *decoder)
1798 {
1799 uint64_t timestamp;
1800 uint32_t mtc, mtc_delta;
1801
1802 if (!decoder->have_tma)
1803 return;
1804
1805 mtc = decoder->packet.payload;
1806
1807 if (decoder->mtc_shift > 8 && decoder->fixup_last_mtc) {
1808 decoder->fixup_last_mtc = false;
1809 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
1810 &decoder->last_mtc);
1811 }
1812
1813 if (mtc > decoder->last_mtc)
1814 mtc_delta = mtc - decoder->last_mtc;
1815 else
1816 mtc_delta = mtc + 256 - decoder->last_mtc;
1817
1818 decoder->ctc_delta += mtc_delta << decoder->mtc_shift;
1819
1820 if (decoder->tsc_ctc_mult) {
1821 timestamp = decoder->ctc_timestamp +
1822 decoder->ctc_delta * decoder->tsc_ctc_mult;
1823 } else {
1824 timestamp = decoder->ctc_timestamp +
1825 multdiv(decoder->ctc_delta,
1826 decoder->tsc_ctc_ratio_n,
1827 decoder->tsc_ctc_ratio_d);
1828 }
1829
1830 if (timestamp < decoder->timestamp)
1831 intel_pt_log("Suppressing MTC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1832 timestamp, decoder->timestamp);
1833 else
1834 decoder->timestamp = timestamp;
1835
1836 intel_pt_mtc_cyc_cnt_upd(decoder);
1837
1838 decoder->timestamp_insn_cnt = 0;
1839 decoder->last_mtc = mtc;
1840
1841 if (decoder->last_packet_type == INTEL_PT_CYC) {
1842 decoder->cyc_ref_timestamp = decoder->timestamp;
1843 decoder->cycle_cnt = 0;
1844 decoder->have_calc_cyc_to_tsc = false;
1845 intel_pt_calc_cyc_to_tsc(decoder, true);
1846 }
1847
1848 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1849 }
1850
intel_pt_calc_cbr(struct intel_pt_decoder * decoder)1851 static void intel_pt_calc_cbr(struct intel_pt_decoder *decoder)
1852 {
1853 unsigned int cbr = decoder->packet.payload & 0xff;
1854
1855 decoder->cbr_payload = decoder->packet.payload;
1856
1857 if (decoder->cbr == cbr)
1858 return;
1859
1860 decoder->cbr = cbr;
1861 decoder->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
1862 decoder->cyc_ref_timestamp = decoder->timestamp;
1863 decoder->cycle_cnt = 0;
1864
1865 intel_pt_mtc_cyc_cnt_cbr(decoder);
1866 }
1867
intel_pt_calc_cyc_timestamp(struct intel_pt_decoder * decoder)1868 static void intel_pt_calc_cyc_timestamp(struct intel_pt_decoder *decoder)
1869 {
1870 uint64_t timestamp = decoder->cyc_ref_timestamp;
1871
1872 decoder->have_cyc = true;
1873
1874 decoder->cycle_cnt += decoder->packet.payload;
1875 if (decoder->pge)
1876 decoder->tot_cyc_cnt += decoder->packet.payload;
1877 decoder->sample_cyc = true;
1878
1879 if (!decoder->cyc_ref_timestamp)
1880 return;
1881
1882 if (decoder->have_calc_cyc_to_tsc)
1883 timestamp += decoder->cycle_cnt * decoder->calc_cyc_to_tsc;
1884 else if (decoder->cbr)
1885 timestamp += decoder->cycle_cnt * decoder->cbr_cyc_to_tsc;
1886 else
1887 return;
1888
1889 if (timestamp < decoder->timestamp)
1890 intel_pt_log("Suppressing CYC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1891 timestamp, decoder->timestamp);
1892 else
1893 decoder->timestamp = timestamp;
1894
1895 decoder->timestamp_insn_cnt = 0;
1896
1897 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1898 }
1899
intel_pt_bbp(struct intel_pt_decoder * decoder)1900 static void intel_pt_bbp(struct intel_pt_decoder *decoder)
1901 {
1902 if (decoder->prev_pkt_ctx == INTEL_PT_NO_CTX) {
1903 memset(decoder->state.items.mask, 0, sizeof(decoder->state.items.mask));
1904 decoder->state.items.is_32_bit = false;
1905 }
1906 decoder->blk_type = decoder->packet.payload;
1907 decoder->blk_type_pos = intel_pt_blk_type_pos(decoder->blk_type);
1908 if (decoder->blk_type == INTEL_PT_GP_REGS)
1909 decoder->state.items.is_32_bit = decoder->packet.count;
1910 if (decoder->blk_type_pos < 0) {
1911 intel_pt_log("WARNING: Unknown block type %u\n",
1912 decoder->blk_type);
1913 } else if (decoder->state.items.mask[decoder->blk_type_pos]) {
1914 intel_pt_log("WARNING: Duplicate block type %u\n",
1915 decoder->blk_type);
1916 }
1917 }
1918
intel_pt_bip(struct intel_pt_decoder * decoder)1919 static void intel_pt_bip(struct intel_pt_decoder *decoder)
1920 {
1921 uint32_t id = decoder->packet.count;
1922 uint32_t bit = 1 << id;
1923 int pos = decoder->blk_type_pos;
1924
1925 if (pos < 0 || id >= INTEL_PT_BLK_ITEM_ID_CNT) {
1926 intel_pt_log("WARNING: Unknown block item %u type %d\n",
1927 id, decoder->blk_type);
1928 return;
1929 }
1930
1931 if (decoder->state.items.mask[pos] & bit) {
1932 intel_pt_log("WARNING: Duplicate block item %u type %d\n",
1933 id, decoder->blk_type);
1934 }
1935
1936 decoder->state.items.mask[pos] |= bit;
1937 decoder->state.items.val[pos][id] = decoder->packet.payload;
1938 }
1939
1940 /* Walk PSB+ packets when already in sync. */
intel_pt_walk_psbend(struct intel_pt_decoder * decoder)1941 static int intel_pt_walk_psbend(struct intel_pt_decoder *decoder)
1942 {
1943 int err;
1944
1945 decoder->in_psb = true;
1946
1947 while (1) {
1948 err = intel_pt_get_next_packet(decoder);
1949 if (err)
1950 goto out;
1951
1952 switch (decoder->packet.type) {
1953 case INTEL_PT_PSBEND:
1954 err = 0;
1955 goto out;
1956
1957 case INTEL_PT_TIP_PGD:
1958 case INTEL_PT_TIP_PGE:
1959 case INTEL_PT_TIP:
1960 case INTEL_PT_TNT:
1961 case INTEL_PT_TRACESTOP:
1962 case INTEL_PT_BAD:
1963 case INTEL_PT_PSB:
1964 case INTEL_PT_PTWRITE:
1965 case INTEL_PT_PTWRITE_IP:
1966 case INTEL_PT_EXSTOP:
1967 case INTEL_PT_EXSTOP_IP:
1968 case INTEL_PT_MWAIT:
1969 case INTEL_PT_PWRE:
1970 case INTEL_PT_PWRX:
1971 case INTEL_PT_BBP:
1972 case INTEL_PT_BIP:
1973 case INTEL_PT_BEP:
1974 case INTEL_PT_BEP_IP:
1975 case INTEL_PT_CFE:
1976 case INTEL_PT_CFE_IP:
1977 case INTEL_PT_EVD:
1978 decoder->have_tma = false;
1979 intel_pt_log("ERROR: Unexpected packet\n");
1980 err = -EAGAIN;
1981 goto out;
1982
1983 case INTEL_PT_OVF:
1984 err = intel_pt_overflow(decoder);
1985 goto out;
1986
1987 case INTEL_PT_TSC:
1988 intel_pt_calc_tsc_timestamp(decoder);
1989 break;
1990
1991 case INTEL_PT_TMA:
1992 intel_pt_calc_tma(decoder);
1993 break;
1994
1995 case INTEL_PT_CBR:
1996 intel_pt_calc_cbr(decoder);
1997 break;
1998
1999 case INTEL_PT_MODE_EXEC:
2000 decoder->exec_mode = decoder->packet.payload;
2001 break;
2002
2003 case INTEL_PT_PIP:
2004 intel_pt_set_pip(decoder);
2005 break;
2006
2007 case INTEL_PT_FUP:
2008 decoder->pge = true;
2009 if (decoder->packet.count) {
2010 intel_pt_set_last_ip(decoder);
2011 decoder->psb_ip = decoder->last_ip;
2012 }
2013 break;
2014
2015 case INTEL_PT_MODE_TSX:
2016 intel_pt_update_in_tx(decoder);
2017 break;
2018
2019 case INTEL_PT_MTC:
2020 intel_pt_calc_mtc_timestamp(decoder);
2021 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
2022 decoder->state.type |= INTEL_PT_INSTRUCTION;
2023 break;
2024
2025 case INTEL_PT_CYC:
2026 intel_pt_calc_cyc_timestamp(decoder);
2027 break;
2028
2029 case INTEL_PT_VMCS:
2030 case INTEL_PT_MNT:
2031 case INTEL_PT_PAD:
2032 default:
2033 break;
2034 }
2035 }
2036 out:
2037 decoder->in_psb = false;
2038
2039 return err;
2040 }
2041
intel_pt_walk_fup_tip(struct intel_pt_decoder * decoder)2042 static int intel_pt_walk_fup_tip(struct intel_pt_decoder *decoder)
2043 {
2044 int err;
2045
2046 if (decoder->tx_flags & INTEL_PT_ABORT_TX) {
2047 decoder->tx_flags = 0;
2048 decoder->state.flags &= ~INTEL_PT_IN_TX;
2049 decoder->state.flags |= INTEL_PT_ABORT_TX;
2050 } else {
2051 decoder->state.flags |= INTEL_PT_ASYNC;
2052 }
2053
2054 while (1) {
2055 err = intel_pt_get_next_packet(decoder);
2056 if (err)
2057 return err;
2058
2059 switch (decoder->packet.type) {
2060 case INTEL_PT_TNT:
2061 case INTEL_PT_FUP:
2062 case INTEL_PT_TRACESTOP:
2063 case INTEL_PT_PSB:
2064 case INTEL_PT_TSC:
2065 case INTEL_PT_TMA:
2066 case INTEL_PT_MODE_TSX:
2067 case INTEL_PT_BAD:
2068 case INTEL_PT_PSBEND:
2069 case INTEL_PT_PTWRITE:
2070 case INTEL_PT_PTWRITE_IP:
2071 case INTEL_PT_EXSTOP:
2072 case INTEL_PT_EXSTOP_IP:
2073 case INTEL_PT_MWAIT:
2074 case INTEL_PT_PWRE:
2075 case INTEL_PT_PWRX:
2076 case INTEL_PT_BBP:
2077 case INTEL_PT_BIP:
2078 case INTEL_PT_BEP:
2079 case INTEL_PT_BEP_IP:
2080 case INTEL_PT_CFE:
2081 case INTEL_PT_CFE_IP:
2082 case INTEL_PT_EVD:
2083 intel_pt_log("ERROR: Missing TIP after FUP\n");
2084 decoder->pkt_state = INTEL_PT_STATE_ERR3;
2085 decoder->pkt_step = 0;
2086 return -ENOENT;
2087
2088 case INTEL_PT_CBR:
2089 intel_pt_calc_cbr(decoder);
2090 break;
2091
2092 case INTEL_PT_OVF:
2093 return intel_pt_overflow(decoder);
2094
2095 case INTEL_PT_TIP_PGD:
2096 decoder->state.from_ip = decoder->ip;
2097 if (decoder->packet.count == 0) {
2098 decoder->state.to_ip = 0;
2099 } else {
2100 intel_pt_set_ip(decoder);
2101 decoder->state.to_ip = decoder->ip;
2102 }
2103 decoder->pge = false;
2104 decoder->continuous_period = false;
2105 decoder->state.type |= INTEL_PT_TRACE_END;
2106 intel_pt_update_nr(decoder);
2107 return 0;
2108
2109 case INTEL_PT_TIP_PGE:
2110 decoder->pge = true;
2111 intel_pt_log("Omitting PGE ip " x64_fmt "\n",
2112 decoder->ip);
2113 decoder->state.from_ip = 0;
2114 if (decoder->packet.count == 0) {
2115 decoder->state.to_ip = 0;
2116 } else {
2117 intel_pt_set_ip(decoder);
2118 decoder->state.to_ip = decoder->ip;
2119 }
2120 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
2121 intel_pt_mtc_cyc_cnt_pge(decoder);
2122 intel_pt_set_nr(decoder);
2123 return 0;
2124
2125 case INTEL_PT_TIP:
2126 decoder->state.from_ip = decoder->ip;
2127 if (decoder->packet.count == 0) {
2128 decoder->state.to_ip = 0;
2129 } else {
2130 intel_pt_set_ip(decoder);
2131 decoder->state.to_ip = decoder->ip;
2132 }
2133 intel_pt_update_nr(decoder);
2134 return 0;
2135
2136 case INTEL_PT_PIP:
2137 intel_pt_update_pip(decoder);
2138 break;
2139
2140 case INTEL_PT_MTC:
2141 intel_pt_calc_mtc_timestamp(decoder);
2142 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
2143 decoder->state.type |= INTEL_PT_INSTRUCTION;
2144 break;
2145
2146 case INTEL_PT_CYC:
2147 intel_pt_calc_cyc_timestamp(decoder);
2148 break;
2149
2150 case INTEL_PT_MODE_EXEC:
2151 decoder->exec_mode = decoder->packet.payload;
2152 break;
2153
2154 case INTEL_PT_VMCS:
2155 case INTEL_PT_MNT:
2156 case INTEL_PT_PAD:
2157 break;
2158
2159 default:
2160 return intel_pt_bug(decoder);
2161 }
2162 }
2163 }
2164
intel_pt_resample(struct intel_pt_decoder * decoder)2165 static int intel_pt_resample(struct intel_pt_decoder *decoder)
2166 {
2167 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2168 decoder->state.type = INTEL_PT_INSTRUCTION;
2169 decoder->state.from_ip = decoder->ip;
2170 decoder->state.to_ip = 0;
2171 return 0;
2172 }
2173
2174 struct intel_pt_vm_tsc_info {
2175 struct intel_pt_pkt pip_packet;
2176 struct intel_pt_pkt vmcs_packet;
2177 struct intel_pt_pkt tma_packet;
2178 bool tsc, pip, vmcs, tma, psbend;
2179 uint64_t ctc_delta;
2180 uint64_t last_ctc;
2181 int max_lookahead;
2182 };
2183
2184 /* Lookahead and get the PIP, VMCS and TMA packets from PSB+ */
intel_pt_vm_psb_lookahead_cb(struct intel_pt_pkt_info * pkt_info)2185 static int intel_pt_vm_psb_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
2186 {
2187 struct intel_pt_vm_tsc_info *data = pkt_info->data;
2188
2189 switch (pkt_info->packet.type) {
2190 case INTEL_PT_PAD:
2191 case INTEL_PT_MNT:
2192 case INTEL_PT_MODE_EXEC:
2193 case INTEL_PT_MODE_TSX:
2194 case INTEL_PT_MTC:
2195 case INTEL_PT_FUP:
2196 case INTEL_PT_CYC:
2197 case INTEL_PT_CBR:
2198 break;
2199
2200 case INTEL_PT_TSC:
2201 data->tsc = true;
2202 break;
2203
2204 case INTEL_PT_TMA:
2205 data->tma_packet = pkt_info->packet;
2206 data->tma = true;
2207 break;
2208
2209 case INTEL_PT_PIP:
2210 data->pip_packet = pkt_info->packet;
2211 data->pip = true;
2212 break;
2213
2214 case INTEL_PT_VMCS:
2215 data->vmcs_packet = pkt_info->packet;
2216 data->vmcs = true;
2217 break;
2218
2219 case INTEL_PT_PSBEND:
2220 data->psbend = true;
2221 return 1;
2222
2223 case INTEL_PT_TIP_PGE:
2224 case INTEL_PT_PTWRITE:
2225 case INTEL_PT_PTWRITE_IP:
2226 case INTEL_PT_EXSTOP:
2227 case INTEL_PT_EXSTOP_IP:
2228 case INTEL_PT_MWAIT:
2229 case INTEL_PT_PWRE:
2230 case INTEL_PT_PWRX:
2231 case INTEL_PT_BBP:
2232 case INTEL_PT_BIP:
2233 case INTEL_PT_BEP:
2234 case INTEL_PT_BEP_IP:
2235 case INTEL_PT_OVF:
2236 case INTEL_PT_BAD:
2237 case INTEL_PT_TNT:
2238 case INTEL_PT_TIP_PGD:
2239 case INTEL_PT_TIP:
2240 case INTEL_PT_PSB:
2241 case INTEL_PT_TRACESTOP:
2242 case INTEL_PT_CFE:
2243 case INTEL_PT_CFE_IP:
2244 case INTEL_PT_EVD:
2245 default:
2246 return 1;
2247 }
2248
2249 return 0;
2250 }
2251
2252 struct intel_pt_ovf_fup_info {
2253 int max_lookahead;
2254 bool found;
2255 };
2256
2257 /* Lookahead to detect a FUP packet after OVF */
intel_pt_ovf_fup_lookahead_cb(struct intel_pt_pkt_info * pkt_info)2258 static int intel_pt_ovf_fup_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
2259 {
2260 struct intel_pt_ovf_fup_info *data = pkt_info->data;
2261
2262 if (pkt_info->packet.type == INTEL_PT_CYC ||
2263 pkt_info->packet.type == INTEL_PT_MTC ||
2264 pkt_info->packet.type == INTEL_PT_TSC)
2265 return !--(data->max_lookahead);
2266 data->found = pkt_info->packet.type == INTEL_PT_FUP;
2267 return 1;
2268 }
2269
intel_pt_ovf_fup_lookahead(struct intel_pt_decoder * decoder)2270 static bool intel_pt_ovf_fup_lookahead(struct intel_pt_decoder *decoder)
2271 {
2272 struct intel_pt_ovf_fup_info data = {
2273 .max_lookahead = 16,
2274 .found = false,
2275 };
2276
2277 intel_pt_pkt_lookahead(decoder, intel_pt_ovf_fup_lookahead_cb, &data);
2278 return data.found;
2279 }
2280
2281 /* Lookahead and get the TMA packet after TSC */
intel_pt_tma_lookahead_cb(struct intel_pt_pkt_info * pkt_info)2282 static int intel_pt_tma_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
2283 {
2284 struct intel_pt_vm_tsc_info *data = pkt_info->data;
2285
2286 if (pkt_info->packet.type == INTEL_PT_CYC ||
2287 pkt_info->packet.type == INTEL_PT_MTC)
2288 return !--(data->max_lookahead);
2289
2290 if (pkt_info->packet.type == INTEL_PT_TMA) {
2291 data->tma_packet = pkt_info->packet;
2292 data->tma = true;
2293 }
2294 return 1;
2295 }
2296
intel_pt_ctc_to_tsc(struct intel_pt_decoder * decoder,uint64_t ctc)2297 static uint64_t intel_pt_ctc_to_tsc(struct intel_pt_decoder *decoder, uint64_t ctc)
2298 {
2299 if (decoder->tsc_ctc_mult)
2300 return ctc * decoder->tsc_ctc_mult;
2301 else
2302 return multdiv(ctc, decoder->tsc_ctc_ratio_n, decoder->tsc_ctc_ratio_d);
2303 }
2304
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)2305 static uint64_t intel_pt_calc_expected_tsc(struct intel_pt_decoder *decoder,
2306 uint32_t ctc,
2307 uint32_t fc,
2308 uint64_t last_ctc_timestamp,
2309 uint64_t ctc_delta,
2310 uint32_t last_ctc)
2311 {
2312 /* Number of CTC ticks from last_ctc_timestamp to last_mtc */
2313 uint64_t last_mtc_ctc = last_ctc + ctc_delta;
2314 /*
2315 * Number of CTC ticks from there until current TMA packet. We would
2316 * expect last_mtc_ctc to be before ctc, but the TSC packet can slip
2317 * past an MTC, so a sign-extended value is used.
2318 */
2319 uint64_t delta = (int16_t)((uint16_t)ctc - (uint16_t)last_mtc_ctc);
2320 /* Total CTC ticks from last_ctc_timestamp to current TMA packet */
2321 uint64_t new_ctc_delta = ctc_delta + delta;
2322 uint64_t expected_tsc;
2323
2324 /*
2325 * Convert CTC ticks to TSC ticks, add the starting point
2326 * (last_ctc_timestamp) and the fast counter from the TMA packet.
2327 */
2328 expected_tsc = last_ctc_timestamp + intel_pt_ctc_to_tsc(decoder, new_ctc_delta) + fc;
2329
2330 if (intel_pt_enable_logging) {
2331 intel_pt_log_x64(last_mtc_ctc);
2332 intel_pt_log_x32(last_ctc);
2333 intel_pt_log_x64(ctc_delta);
2334 intel_pt_log_x64(delta);
2335 intel_pt_log_x32(ctc);
2336 intel_pt_log_x64(new_ctc_delta);
2337 intel_pt_log_x64(last_ctc_timestamp);
2338 intel_pt_log_x32(fc);
2339 intel_pt_log_x64(intel_pt_ctc_to_tsc(decoder, new_ctc_delta));
2340 intel_pt_log_x64(expected_tsc);
2341 }
2342
2343 return expected_tsc;
2344 }
2345
intel_pt_expected_tsc(struct intel_pt_decoder * decoder,struct intel_pt_vm_tsc_info * data)2346 static uint64_t intel_pt_expected_tsc(struct intel_pt_decoder *decoder,
2347 struct intel_pt_vm_tsc_info *data)
2348 {
2349 uint32_t ctc = data->tma_packet.payload;
2350 uint32_t fc = data->tma_packet.count;
2351
2352 return intel_pt_calc_expected_tsc(decoder, ctc, fc,
2353 decoder->ctc_timestamp,
2354 data->ctc_delta, data->last_ctc);
2355 }
2356
intel_pt_translate_vm_tsc(struct intel_pt_decoder * decoder,struct intel_pt_vmcs_info * vmcs_info)2357 static void intel_pt_translate_vm_tsc(struct intel_pt_decoder *decoder,
2358 struct intel_pt_vmcs_info *vmcs_info)
2359 {
2360 uint64_t payload = decoder->packet.payload;
2361
2362 /* VMX adds the TSC Offset, so subtract to get host TSC */
2363 decoder->packet.payload -= vmcs_info->tsc_offset;
2364 /* TSC packet has only 7 bytes */
2365 decoder->packet.payload &= SEVEN_BYTES;
2366
2367 /*
2368 * The buffer is mmapped from the data file, so this also updates the
2369 * data file.
2370 */
2371 if (!decoder->vm_tm_corr_dry_run)
2372 memcpy((void *)decoder->buf + 1, &decoder->packet.payload, 7);
2373
2374 intel_pt_log("Translated VM TSC %#" PRIx64 " -> %#" PRIx64
2375 " VMCS %#" PRIx64 " TSC Offset %#" PRIx64 "\n",
2376 payload, decoder->packet.payload, vmcs_info->vmcs,
2377 vmcs_info->tsc_offset);
2378 }
2379
intel_pt_translate_vm_tsc_offset(struct intel_pt_decoder * decoder,uint64_t tsc_offset)2380 static void intel_pt_translate_vm_tsc_offset(struct intel_pt_decoder *decoder,
2381 uint64_t tsc_offset)
2382 {
2383 struct intel_pt_vmcs_info vmcs_info = {
2384 .vmcs = NO_VMCS,
2385 .tsc_offset = tsc_offset
2386 };
2387
2388 intel_pt_translate_vm_tsc(decoder, &vmcs_info);
2389 }
2390
in_vm(uint64_t pip_payload)2391 static inline bool in_vm(uint64_t pip_payload)
2392 {
2393 return pip_payload & 1;
2394 }
2395
pip_in_vm(struct intel_pt_pkt * pip_packet)2396 static inline bool pip_in_vm(struct intel_pt_pkt *pip_packet)
2397 {
2398 return pip_packet->payload & 1;
2399 }
2400
intel_pt_print_vmcs_info(struct intel_pt_vmcs_info * vmcs_info)2401 static void intel_pt_print_vmcs_info(struct intel_pt_vmcs_info *vmcs_info)
2402 {
2403 p_log("VMCS: %#" PRIx64 " TSC Offset %#" PRIx64,
2404 vmcs_info->vmcs, vmcs_info->tsc_offset);
2405 }
2406
intel_pt_vm_tm_corr_psb(struct intel_pt_decoder * decoder,struct intel_pt_vm_tsc_info * data)2407 static void intel_pt_vm_tm_corr_psb(struct intel_pt_decoder *decoder,
2408 struct intel_pt_vm_tsc_info *data)
2409 {
2410 memset(data, 0, sizeof(*data));
2411 data->ctc_delta = decoder->ctc_delta;
2412 data->last_ctc = decoder->last_ctc;
2413 intel_pt_pkt_lookahead(decoder, intel_pt_vm_psb_lookahead_cb, data);
2414 if (data->tsc && !data->psbend)
2415 p_log("ERROR: PSB without PSBEND");
2416 decoder->in_psb = data->psbend;
2417 }
2418
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)2419 static void intel_pt_vm_tm_corr_first_tsc(struct intel_pt_decoder *decoder,
2420 struct intel_pt_vm_tsc_info *data,
2421 struct intel_pt_vmcs_info *vmcs_info,
2422 uint64_t host_tsc)
2423 {
2424 if (!decoder->in_psb) {
2425 /* Can't happen */
2426 p_log("ERROR: First TSC is not in PSB+");
2427 }
2428
2429 if (data->pip) {
2430 if (pip_in_vm(&data->pip_packet)) { /* Guest */
2431 if (vmcs_info && vmcs_info->tsc_offset) {
2432 intel_pt_translate_vm_tsc(decoder, vmcs_info);
2433 decoder->vm_tm_corr_reliable = true;
2434 } else {
2435 p_log("ERROR: First TSC, unknown TSC Offset");
2436 }
2437 } else { /* Host */
2438 decoder->vm_tm_corr_reliable = true;
2439 }
2440 } else { /* Host or Guest */
2441 decoder->vm_tm_corr_reliable = false;
2442 if (intel_pt_time_in_range(decoder, host_tsc)) {
2443 /* Assume Host */
2444 } else {
2445 /* Assume Guest */
2446 if (vmcs_info && vmcs_info->tsc_offset)
2447 intel_pt_translate_vm_tsc(decoder, vmcs_info);
2448 else
2449 p_log("ERROR: First TSC, no PIP, unknown TSC Offset");
2450 }
2451 }
2452 }
2453
intel_pt_vm_tm_corr_tsc(struct intel_pt_decoder * decoder,struct intel_pt_vm_tsc_info * data)2454 static void intel_pt_vm_tm_corr_tsc(struct intel_pt_decoder *decoder,
2455 struct intel_pt_vm_tsc_info *data)
2456 {
2457 struct intel_pt_vmcs_info *vmcs_info;
2458 uint64_t tsc_offset = 0;
2459 uint64_t vmcs;
2460 bool reliable = true;
2461 uint64_t expected_tsc;
2462 uint64_t host_tsc;
2463 uint64_t ref_timestamp;
2464
2465 bool assign = false;
2466 bool assign_reliable = false;
2467
2468 /* Already have 'data' for the in_psb case */
2469 if (!decoder->in_psb) {
2470 memset(data, 0, sizeof(*data));
2471 data->ctc_delta = decoder->ctc_delta;
2472 data->last_ctc = decoder->last_ctc;
2473 data->max_lookahead = 16;
2474 intel_pt_pkt_lookahead(decoder, intel_pt_tma_lookahead_cb, data);
2475 if (decoder->pge) {
2476 data->pip = true;
2477 data->pip_packet.payload = decoder->pip_payload;
2478 }
2479 }
2480
2481 /* Calculations depend on having TMA packets */
2482 if (!data->tma) {
2483 p_log("ERROR: TSC without TMA");
2484 return;
2485 }
2486
2487 vmcs = data->vmcs ? data->vmcs_packet.payload : decoder->vmcs;
2488 if (vmcs == NO_VMCS)
2489 vmcs = 0;
2490
2491 vmcs_info = decoder->findnew_vmcs_info(decoder->data, vmcs);
2492
2493 ref_timestamp = decoder->timestamp ? decoder->timestamp : decoder->buf_timestamp;
2494 host_tsc = intel_pt_8b_tsc(decoder->packet.payload, ref_timestamp);
2495
2496 if (!decoder->ctc_timestamp) {
2497 intel_pt_vm_tm_corr_first_tsc(decoder, data, vmcs_info, host_tsc);
2498 return;
2499 }
2500
2501 expected_tsc = intel_pt_expected_tsc(decoder, data);
2502
2503 tsc_offset = host_tsc - expected_tsc;
2504
2505 /* Determine if TSC is from Host or Guest */
2506 if (data->pip) {
2507 if (pip_in_vm(&data->pip_packet)) { /* Guest */
2508 if (!vmcs_info) {
2509 /* PIP NR=1 without VMCS cannot happen */
2510 p_log("ERROR: Missing VMCS");
2511 intel_pt_translate_vm_tsc_offset(decoder, tsc_offset);
2512 decoder->vm_tm_corr_reliable = false;
2513 return;
2514 }
2515 } else { /* Host */
2516 decoder->last_reliable_timestamp = host_tsc;
2517 decoder->vm_tm_corr_reliable = true;
2518 return;
2519 }
2520 } else { /* Host or Guest */
2521 reliable = false; /* Host/Guest is a guess, so not reliable */
2522 if (decoder->in_psb) {
2523 if (!tsc_offset)
2524 return; /* Zero TSC Offset, assume Host */
2525 /*
2526 * TSC packet has only 7 bytes of TSC. We have no
2527 * information about the Guest's 8th byte, but it
2528 * doesn't matter because we only need 7 bytes.
2529 * Here, since the 8th byte is unreliable and
2530 * irrelevant, compare only 7 byes.
2531 */
2532 if (vmcs_info &&
2533 (tsc_offset & SEVEN_BYTES) ==
2534 (vmcs_info->tsc_offset & SEVEN_BYTES)) {
2535 /* Same TSC Offset as last VMCS, assume Guest */
2536 goto guest;
2537 }
2538 }
2539 /*
2540 * Check if the host_tsc is within the expected range.
2541 * Note, we could narrow the range more by looking ahead for
2542 * the next host TSC in the same buffer, but we don't bother to
2543 * do that because this is probably good enough.
2544 */
2545 if (host_tsc >= expected_tsc && intel_pt_time_in_range(decoder, host_tsc)) {
2546 /* Within expected range for Host TSC, assume Host */
2547 decoder->vm_tm_corr_reliable = false;
2548 return;
2549 }
2550 }
2551
2552 guest: /* Assuming Guest */
2553
2554 /* Determine whether to assign TSC Offset */
2555 if (vmcs_info && vmcs_info->vmcs) {
2556 if (vmcs_info->tsc_offset && vmcs_info->reliable) {
2557 assign = false;
2558 } else if (decoder->in_psb && data->pip && decoder->vm_tm_corr_reliable &&
2559 decoder->vm_tm_corr_continuous && decoder->vm_tm_corr_same_buf) {
2560 /* Continuous tracing, TSC in a PSB is not a time loss */
2561 assign = true;
2562 assign_reliable = true;
2563 } else if (decoder->in_psb && data->pip && decoder->vm_tm_corr_same_buf) {
2564 /*
2565 * Unlikely to be a time loss TSC in a PSB which is not
2566 * at the start of a buffer.
2567 */
2568 assign = true;
2569 assign_reliable = false;
2570 }
2571 }
2572
2573 /* Record VMCS TSC Offset */
2574 if (assign && (vmcs_info->tsc_offset != tsc_offset ||
2575 vmcs_info->reliable != assign_reliable)) {
2576 bool print = vmcs_info->tsc_offset != tsc_offset;
2577
2578 vmcs_info->tsc_offset = tsc_offset;
2579 vmcs_info->reliable = assign_reliable;
2580 if (print)
2581 intel_pt_print_vmcs_info(vmcs_info);
2582 }
2583
2584 /* Determine what TSC Offset to use */
2585 if (vmcs_info && vmcs_info->tsc_offset) {
2586 if (!vmcs_info->reliable)
2587 reliable = false;
2588 intel_pt_translate_vm_tsc(decoder, vmcs_info);
2589 } else {
2590 reliable = false;
2591 if (vmcs_info) {
2592 if (!vmcs_info->error_printed) {
2593 p_log("ERROR: Unknown TSC Offset for VMCS %#" PRIx64,
2594 vmcs_info->vmcs);
2595 vmcs_info->error_printed = true;
2596 }
2597 } else {
2598 if (intel_pt_print_once(decoder, INTEL_PT_PRT_ONCE_UNK_VMCS))
2599 p_log("ERROR: Unknown VMCS");
2600 }
2601 intel_pt_translate_vm_tsc_offset(decoder, tsc_offset);
2602 }
2603
2604 decoder->vm_tm_corr_reliable = reliable;
2605 }
2606
intel_pt_vm_tm_corr_pebs_tsc(struct intel_pt_decoder * decoder)2607 static void intel_pt_vm_tm_corr_pebs_tsc(struct intel_pt_decoder *decoder)
2608 {
2609 uint64_t host_tsc = decoder->packet.payload;
2610 uint64_t guest_tsc = decoder->packet.payload;
2611 struct intel_pt_vmcs_info *vmcs_info;
2612 uint64_t vmcs;
2613
2614 vmcs = decoder->vmcs;
2615 if (vmcs == NO_VMCS)
2616 vmcs = 0;
2617
2618 vmcs_info = decoder->findnew_vmcs_info(decoder->data, vmcs);
2619
2620 if (decoder->pge) {
2621 if (in_vm(decoder->pip_payload)) { /* Guest */
2622 if (!vmcs_info) {
2623 /* PIP NR=1 without VMCS cannot happen */
2624 p_log("ERROR: Missing VMCS");
2625 }
2626 } else { /* Host */
2627 return;
2628 }
2629 } else { /* Host or Guest */
2630 if (intel_pt_time_in_range(decoder, host_tsc)) {
2631 /* Within expected range for Host TSC, assume Host */
2632 return;
2633 }
2634 }
2635
2636 if (vmcs_info) {
2637 /* Translate Guest TSC to Host TSC */
2638 host_tsc = ((guest_tsc & SEVEN_BYTES) - vmcs_info->tsc_offset) & SEVEN_BYTES;
2639 host_tsc = intel_pt_8b_tsc(host_tsc, decoder->timestamp);
2640 intel_pt_log("Translated VM TSC %#" PRIx64 " -> %#" PRIx64
2641 " VMCS %#" PRIx64 " TSC Offset %#" PRIx64 "\n",
2642 guest_tsc, host_tsc, vmcs_info->vmcs,
2643 vmcs_info->tsc_offset);
2644 if (!intel_pt_time_in_range(decoder, host_tsc) &&
2645 intel_pt_print_once(decoder, INTEL_PT_PRT_ONCE_ERANGE))
2646 p_log("Timestamp out of range");
2647 } else {
2648 if (intel_pt_print_once(decoder, INTEL_PT_PRT_ONCE_UNK_VMCS))
2649 p_log("ERROR: Unknown VMCS");
2650 host_tsc = decoder->timestamp;
2651 }
2652
2653 decoder->packet.payload = host_tsc;
2654
2655 if (!decoder->vm_tm_corr_dry_run)
2656 memcpy((void *)decoder->buf + 1, &host_tsc, 8);
2657 }
2658
intel_pt_vm_time_correlation(struct intel_pt_decoder * decoder)2659 static int intel_pt_vm_time_correlation(struct intel_pt_decoder *decoder)
2660 {
2661 struct intel_pt_vm_tsc_info data = { .psbend = false };
2662 bool pge;
2663 int err;
2664
2665 if (decoder->in_psb)
2666 intel_pt_vm_tm_corr_psb(decoder, &data);
2667
2668 while (1) {
2669 err = intel_pt_get_next_packet(decoder);
2670 if (err == -ENOLINK)
2671 continue;
2672 if (err)
2673 break;
2674
2675 switch (decoder->packet.type) {
2676 case INTEL_PT_TIP_PGD:
2677 decoder->pge = false;
2678 decoder->vm_tm_corr_continuous = false;
2679 break;
2680
2681 case INTEL_PT_TNT:
2682 case INTEL_PT_TIP:
2683 case INTEL_PT_TIP_PGE:
2684 decoder->pge = true;
2685 break;
2686
2687 case INTEL_PT_OVF:
2688 decoder->in_psb = false;
2689 pge = decoder->pge;
2690 decoder->pge = intel_pt_ovf_fup_lookahead(decoder);
2691 if (pge != decoder->pge)
2692 intel_pt_log("Surprising PGE change in OVF!");
2693 if (!decoder->pge)
2694 decoder->vm_tm_corr_continuous = false;
2695 break;
2696
2697 case INTEL_PT_FUP:
2698 if (decoder->in_psb)
2699 decoder->pge = true;
2700 break;
2701
2702 case INTEL_PT_TRACESTOP:
2703 decoder->pge = false;
2704 decoder->vm_tm_corr_continuous = false;
2705 decoder->have_tma = false;
2706 break;
2707
2708 case INTEL_PT_PSB:
2709 intel_pt_vm_tm_corr_psb(decoder, &data);
2710 break;
2711
2712 case INTEL_PT_PIP:
2713 decoder->pip_payload = decoder->packet.payload;
2714 break;
2715
2716 case INTEL_PT_MTC:
2717 intel_pt_calc_mtc_timestamp(decoder);
2718 break;
2719
2720 case INTEL_PT_TSC:
2721 intel_pt_vm_tm_corr_tsc(decoder, &data);
2722 intel_pt_calc_tsc_timestamp(decoder);
2723 decoder->vm_tm_corr_same_buf = true;
2724 decoder->vm_tm_corr_continuous = decoder->pge;
2725 break;
2726
2727 case INTEL_PT_TMA:
2728 intel_pt_calc_tma(decoder);
2729 break;
2730
2731 case INTEL_PT_CYC:
2732 intel_pt_calc_cyc_timestamp(decoder);
2733 break;
2734
2735 case INTEL_PT_CBR:
2736 intel_pt_calc_cbr(decoder);
2737 break;
2738
2739 case INTEL_PT_PSBEND:
2740 decoder->in_psb = false;
2741 data.psbend = false;
2742 break;
2743
2744 case INTEL_PT_VMCS:
2745 if (decoder->packet.payload != NO_VMCS)
2746 decoder->vmcs = decoder->packet.payload;
2747 break;
2748
2749 case INTEL_PT_BBP:
2750 decoder->blk_type = decoder->packet.payload;
2751 break;
2752
2753 case INTEL_PT_BIP:
2754 if (decoder->blk_type == INTEL_PT_PEBS_BASIC &&
2755 decoder->packet.count == 2)
2756 intel_pt_vm_tm_corr_pebs_tsc(decoder);
2757 break;
2758
2759 case INTEL_PT_BEP:
2760 case INTEL_PT_BEP_IP:
2761 decoder->blk_type = 0;
2762 break;
2763
2764 case INTEL_PT_CFE:
2765 case INTEL_PT_CFE_IP:
2766 case INTEL_PT_EVD:
2767 case INTEL_PT_MODE_EXEC:
2768 case INTEL_PT_MODE_TSX:
2769 case INTEL_PT_MNT:
2770 case INTEL_PT_PAD:
2771 case INTEL_PT_PTWRITE_IP:
2772 case INTEL_PT_PTWRITE:
2773 case INTEL_PT_MWAIT:
2774 case INTEL_PT_PWRE:
2775 case INTEL_PT_EXSTOP_IP:
2776 case INTEL_PT_EXSTOP:
2777 case INTEL_PT_PWRX:
2778 case INTEL_PT_BAD: /* Does not happen */
2779 default:
2780 break;
2781 }
2782 }
2783
2784 return err;
2785 }
2786
2787 #define HOP_PROCESS 0
2788 #define HOP_IGNORE 1
2789 #define HOP_RETURN 2
2790 #define HOP_AGAIN 3
2791
2792 static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder);
2793
2794 /* 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)2795 static int intel_pt_hop_trace(struct intel_pt_decoder *decoder, bool *no_tip, int *err)
2796 {
2797 *err = 0;
2798
2799 /* Leap from PSB to PSB, getting ip from FUP within PSB+ */
2800 if (decoder->leap && !decoder->in_psb && decoder->packet.type != INTEL_PT_PSB) {
2801 *err = intel_pt_scan_for_psb(decoder);
2802 if (*err)
2803 return HOP_RETURN;
2804 }
2805
2806 switch (decoder->packet.type) {
2807 case INTEL_PT_TNT:
2808 return HOP_IGNORE;
2809
2810 case INTEL_PT_TIP_PGD:
2811 decoder->pge = false;
2812 if (!decoder->packet.count) {
2813 intel_pt_set_nr(decoder);
2814 return HOP_IGNORE;
2815 }
2816 intel_pt_set_ip(decoder);
2817 decoder->state.type |= INTEL_PT_TRACE_END;
2818 decoder->state.from_ip = 0;
2819 decoder->state.to_ip = decoder->ip;
2820 intel_pt_update_nr(decoder);
2821 return HOP_RETURN;
2822
2823 case INTEL_PT_TIP:
2824 if (!decoder->packet.count) {
2825 intel_pt_set_nr(decoder);
2826 return HOP_IGNORE;
2827 }
2828 intel_pt_set_ip(decoder);
2829 decoder->state.type = INTEL_PT_INSTRUCTION;
2830 decoder->state.from_ip = decoder->ip;
2831 decoder->state.to_ip = 0;
2832 intel_pt_update_nr(decoder);
2833 return HOP_RETURN;
2834
2835 case INTEL_PT_FUP:
2836 if (!decoder->packet.count)
2837 return HOP_IGNORE;
2838 intel_pt_set_ip(decoder);
2839 if (decoder->set_fup_mwait || decoder->set_fup_pwre)
2840 *no_tip = true;
2841 if (!decoder->branch_enable || !decoder->pge)
2842 *no_tip = true;
2843 if (*no_tip) {
2844 decoder->state.type = INTEL_PT_INSTRUCTION;
2845 decoder->state.from_ip = decoder->ip;
2846 decoder->state.to_ip = 0;
2847 intel_pt_fup_event(decoder);
2848 return HOP_RETURN;
2849 }
2850 intel_pt_fup_event(decoder);
2851 decoder->state.type |= INTEL_PT_INSTRUCTION | INTEL_PT_BRANCH;
2852 *err = intel_pt_walk_fup_tip(decoder);
2853 if (!*err && decoder->state.to_ip)
2854 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
2855 return HOP_RETURN;
2856
2857 case INTEL_PT_PSB:
2858 decoder->state.psb_offset = decoder->pos;
2859 decoder->psb_ip = 0;
2860 decoder->last_ip = 0;
2861 decoder->have_last_ip = true;
2862 *err = intel_pt_walk_psbend(decoder);
2863 if (*err == -EAGAIN)
2864 return HOP_AGAIN;
2865 if (*err)
2866 return HOP_RETURN;
2867 decoder->state.type = INTEL_PT_PSB_EVT;
2868 if (decoder->psb_ip) {
2869 decoder->state.type |= INTEL_PT_INSTRUCTION;
2870 decoder->ip = decoder->psb_ip;
2871 }
2872 decoder->state.from_ip = decoder->psb_ip;
2873 decoder->state.to_ip = 0;
2874 return HOP_RETURN;
2875
2876 case INTEL_PT_BAD:
2877 case INTEL_PT_PAD:
2878 case INTEL_PT_TIP_PGE:
2879 case INTEL_PT_TSC:
2880 case INTEL_PT_TMA:
2881 case INTEL_PT_MODE_EXEC:
2882 case INTEL_PT_MODE_TSX:
2883 case INTEL_PT_MTC:
2884 case INTEL_PT_CYC:
2885 case INTEL_PT_VMCS:
2886 case INTEL_PT_PSBEND:
2887 case INTEL_PT_CBR:
2888 case INTEL_PT_TRACESTOP:
2889 case INTEL_PT_PIP:
2890 case INTEL_PT_OVF:
2891 case INTEL_PT_MNT:
2892 case INTEL_PT_PTWRITE:
2893 case INTEL_PT_PTWRITE_IP:
2894 case INTEL_PT_EXSTOP:
2895 case INTEL_PT_EXSTOP_IP:
2896 case INTEL_PT_MWAIT:
2897 case INTEL_PT_PWRE:
2898 case INTEL_PT_PWRX:
2899 case INTEL_PT_BBP:
2900 case INTEL_PT_BIP:
2901 case INTEL_PT_BEP:
2902 case INTEL_PT_BEP_IP:
2903 case INTEL_PT_CFE:
2904 case INTEL_PT_CFE_IP:
2905 case INTEL_PT_EVD:
2906 default:
2907 return HOP_PROCESS;
2908 }
2909 }
2910
2911 struct intel_pt_psb_info {
2912 struct intel_pt_pkt fup_packet;
2913 bool fup;
2914 int after_psbend;
2915 };
2916
2917 /* Lookahead and get the FUP packet from PSB+ */
intel_pt_psb_lookahead_cb(struct intel_pt_pkt_info * pkt_info)2918 static int intel_pt_psb_lookahead_cb(struct intel_pt_pkt_info *pkt_info)
2919 {
2920 struct intel_pt_psb_info *data = pkt_info->data;
2921
2922 switch (pkt_info->packet.type) {
2923 case INTEL_PT_PAD:
2924 case INTEL_PT_MNT:
2925 case INTEL_PT_TSC:
2926 case INTEL_PT_TMA:
2927 case INTEL_PT_MODE_EXEC:
2928 case INTEL_PT_MODE_TSX:
2929 case INTEL_PT_MTC:
2930 case INTEL_PT_CYC:
2931 case INTEL_PT_VMCS:
2932 case INTEL_PT_CBR:
2933 case INTEL_PT_PIP:
2934 if (data->after_psbend) {
2935 data->after_psbend -= 1;
2936 if (!data->after_psbend)
2937 return 1;
2938 }
2939 break;
2940
2941 case INTEL_PT_FUP:
2942 if (data->after_psbend)
2943 return 1;
2944 if (data->fup || pkt_info->packet.count == 0)
2945 return 1;
2946 data->fup_packet = pkt_info->packet;
2947 data->fup = true;
2948 break;
2949
2950 case INTEL_PT_PSBEND:
2951 if (!data->fup)
2952 return 1;
2953 /* Keep going to check for a TIP.PGE */
2954 data->after_psbend = 6;
2955 break;
2956
2957 case INTEL_PT_TIP_PGE:
2958 /* Ignore FUP in PSB+ if followed by TIP.PGE */
2959 if (data->after_psbend)
2960 data->fup = false;
2961 return 1;
2962
2963 case INTEL_PT_PTWRITE:
2964 case INTEL_PT_PTWRITE_IP:
2965 case INTEL_PT_EXSTOP:
2966 case INTEL_PT_EXSTOP_IP:
2967 case INTEL_PT_MWAIT:
2968 case INTEL_PT_PWRE:
2969 case INTEL_PT_PWRX:
2970 case INTEL_PT_BBP:
2971 case INTEL_PT_BIP:
2972 case INTEL_PT_BEP:
2973 case INTEL_PT_BEP_IP:
2974 case INTEL_PT_CFE:
2975 case INTEL_PT_CFE_IP:
2976 case INTEL_PT_EVD:
2977 if (data->after_psbend) {
2978 data->after_psbend -= 1;
2979 if (!data->after_psbend)
2980 return 1;
2981 break;
2982 }
2983 return 1;
2984
2985 case INTEL_PT_OVF:
2986 case INTEL_PT_BAD:
2987 case INTEL_PT_TNT:
2988 case INTEL_PT_TIP_PGD:
2989 case INTEL_PT_TIP:
2990 case INTEL_PT_PSB:
2991 case INTEL_PT_TRACESTOP:
2992 default:
2993 return 1;
2994 }
2995
2996 return 0;
2997 }
2998
intel_pt_psb(struct intel_pt_decoder * decoder)2999 static int intel_pt_psb(struct intel_pt_decoder *decoder)
3000 {
3001 int err;
3002
3003 decoder->last_ip = 0;
3004 decoder->psb_ip = 0;
3005 decoder->have_last_ip = true;
3006 intel_pt_clear_stack(&decoder->stack);
3007 err = intel_pt_walk_psbend(decoder);
3008 if (err)
3009 return err;
3010 decoder->state.type = INTEL_PT_PSB_EVT;
3011 decoder->state.from_ip = decoder->psb_ip;
3012 decoder->state.to_ip = 0;
3013 return 0;
3014 }
3015
intel_pt_fup_in_psb(struct intel_pt_decoder * decoder)3016 static int intel_pt_fup_in_psb(struct intel_pt_decoder *decoder)
3017 {
3018 int err;
3019
3020 if (decoder->ip != decoder->last_ip) {
3021 err = intel_pt_walk_fup(decoder);
3022 if (!err || err != -EAGAIN)
3023 return err;
3024 }
3025
3026 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3027 err = intel_pt_psb(decoder);
3028 if (err) {
3029 decoder->pkt_state = INTEL_PT_STATE_ERR3;
3030 return -ENOENT;
3031 }
3032
3033 return 0;
3034 }
3035
intel_pt_psb_with_fup(struct intel_pt_decoder * decoder,int * err)3036 static bool intel_pt_psb_with_fup(struct intel_pt_decoder *decoder, int *err)
3037 {
3038 struct intel_pt_psb_info data = { .fup = false };
3039
3040 if (!decoder->branch_enable)
3041 return false;
3042
3043 intel_pt_pkt_lookahead(decoder, intel_pt_psb_lookahead_cb, &data);
3044 if (!data.fup)
3045 return false;
3046
3047 decoder->packet = data.fup_packet;
3048 intel_pt_set_last_ip(decoder);
3049 decoder->pkt_state = INTEL_PT_STATE_FUP_IN_PSB;
3050
3051 *err = intel_pt_fup_in_psb(decoder);
3052
3053 return true;
3054 }
3055
intel_pt_walk_trace(struct intel_pt_decoder * decoder)3056 static int intel_pt_walk_trace(struct intel_pt_decoder *decoder)
3057 {
3058 int last_packet_type = INTEL_PT_PAD;
3059 bool no_tip = false;
3060 int err;
3061
3062 while (1) {
3063 err = intel_pt_get_next_packet(decoder);
3064 if (err)
3065 return err;
3066 next:
3067 err = 0;
3068 if (decoder->cyc_threshold) {
3069 if (decoder->sample_cyc && last_packet_type != INTEL_PT_CYC)
3070 decoder->sample_cyc = false;
3071 last_packet_type = decoder->packet.type;
3072 }
3073
3074 if (decoder->hop) {
3075 switch (intel_pt_hop_trace(decoder, &no_tip, &err)) {
3076 case HOP_IGNORE:
3077 continue;
3078 case HOP_RETURN:
3079 return err;
3080 case HOP_AGAIN:
3081 goto next;
3082 default:
3083 break;
3084 }
3085 }
3086
3087 switch (decoder->packet.type) {
3088 case INTEL_PT_TNT:
3089 if (!decoder->packet.count)
3090 break;
3091 decoder->tnt = decoder->packet;
3092 decoder->pkt_state = INTEL_PT_STATE_TNT;
3093 err = intel_pt_walk_tnt(decoder);
3094 if (err == -EAGAIN)
3095 break;
3096 return err;
3097
3098 case INTEL_PT_TIP_PGD:
3099 if (decoder->packet.count != 0)
3100 intel_pt_set_last_ip(decoder);
3101 decoder->pkt_state = INTEL_PT_STATE_TIP_PGD;
3102 return intel_pt_walk_tip(decoder);
3103
3104 case INTEL_PT_TIP_PGE: {
3105 decoder->pge = true;
3106 decoder->overflow = false;
3107 intel_pt_mtc_cyc_cnt_pge(decoder);
3108 intel_pt_set_nr(decoder);
3109 if (decoder->packet.count == 0) {
3110 intel_pt_log_at("Skipping zero TIP.PGE",
3111 decoder->pos);
3112 break;
3113 }
3114 intel_pt_set_ip(decoder);
3115 decoder->state.from_ip = 0;
3116 decoder->state.to_ip = decoder->ip;
3117 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
3118 /*
3119 * In hop mode, resample to get the to_ip as an
3120 * "instruction" sample.
3121 */
3122 if (decoder->hop)
3123 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
3124 return 0;
3125 }
3126
3127 case INTEL_PT_OVF:
3128 return intel_pt_overflow(decoder);
3129
3130 case INTEL_PT_TIP:
3131 if (decoder->packet.count != 0)
3132 intel_pt_set_last_ip(decoder);
3133 decoder->pkt_state = INTEL_PT_STATE_TIP;
3134 return intel_pt_walk_tip(decoder);
3135
3136 case INTEL_PT_FUP:
3137 if (decoder->packet.count == 0) {
3138 intel_pt_log_at("Skipping zero FUP",
3139 decoder->pos);
3140 no_tip = false;
3141 break;
3142 }
3143 intel_pt_set_last_ip(decoder);
3144 if (!decoder->branch_enable || !decoder->pge) {
3145 decoder->ip = decoder->last_ip;
3146 if (intel_pt_fup_event(decoder))
3147 return 0;
3148 no_tip = false;
3149 break;
3150 }
3151 if (decoder->set_fup_mwait)
3152 no_tip = true;
3153 if (no_tip)
3154 decoder->pkt_state = INTEL_PT_STATE_FUP_NO_TIP;
3155 else
3156 decoder->pkt_state = INTEL_PT_STATE_FUP;
3157 err = intel_pt_walk_fup(decoder);
3158 if (err != -EAGAIN)
3159 return err;
3160 if (no_tip) {
3161 no_tip = false;
3162 break;
3163 }
3164 return intel_pt_walk_fup_tip(decoder);
3165
3166 case INTEL_PT_TRACESTOP:
3167 decoder->pge = false;
3168 decoder->continuous_period = false;
3169 intel_pt_clear_tx_flags(decoder);
3170 decoder->have_tma = false;
3171 break;
3172
3173 case INTEL_PT_PSB:
3174 decoder->state.psb_offset = decoder->pos;
3175 decoder->psb_ip = 0;
3176 if (intel_pt_psb_with_fup(decoder, &err))
3177 return err;
3178 err = intel_pt_psb(decoder);
3179 if (err == -EAGAIN)
3180 goto next;
3181 return err;
3182
3183 case INTEL_PT_PIP:
3184 intel_pt_update_pip(decoder);
3185 break;
3186
3187 case INTEL_PT_MTC:
3188 intel_pt_calc_mtc_timestamp(decoder);
3189 if (decoder->period_type != INTEL_PT_PERIOD_MTC)
3190 break;
3191 /*
3192 * Ensure that there has been an instruction since the
3193 * last MTC.
3194 */
3195 if (!decoder->mtc_insn)
3196 break;
3197 decoder->mtc_insn = false;
3198 /* Ensure that there is a timestamp */
3199 if (!decoder->timestamp)
3200 break;
3201 decoder->state.type = INTEL_PT_INSTRUCTION;
3202 decoder->state.from_ip = decoder->ip;
3203 decoder->state.to_ip = 0;
3204 decoder->mtc_insn = false;
3205 return 0;
3206
3207 case INTEL_PT_TSC:
3208 intel_pt_calc_tsc_timestamp(decoder);
3209 break;
3210
3211 case INTEL_PT_TMA:
3212 intel_pt_calc_tma(decoder);
3213 break;
3214
3215 case INTEL_PT_CYC:
3216 intel_pt_calc_cyc_timestamp(decoder);
3217 break;
3218
3219 case INTEL_PT_CBR:
3220 intel_pt_calc_cbr(decoder);
3221 if (decoder->cbr != decoder->cbr_seen) {
3222 decoder->state.type = 0;
3223 return 0;
3224 }
3225 break;
3226
3227 case INTEL_PT_MODE_EXEC:
3228 decoder->exec_mode = decoder->packet.payload;
3229 break;
3230
3231 case INTEL_PT_MODE_TSX:
3232 /* MODE_TSX need not be followed by FUP */
3233 if (!decoder->pge || decoder->in_psb) {
3234 intel_pt_update_in_tx(decoder);
3235 break;
3236 }
3237 err = intel_pt_mode_tsx(decoder, &no_tip);
3238 if (err)
3239 return err;
3240 goto next;
3241
3242 case INTEL_PT_BAD: /* Does not happen */
3243 return intel_pt_bug(decoder);
3244
3245 case INTEL_PT_PSBEND:
3246 case INTEL_PT_VMCS:
3247 case INTEL_PT_MNT:
3248 case INTEL_PT_PAD:
3249 break;
3250
3251 case INTEL_PT_PTWRITE_IP:
3252 decoder->fup_ptw_payload = decoder->packet.payload;
3253 err = intel_pt_get_next_packet(decoder);
3254 if (err)
3255 return err;
3256 if (decoder->packet.type == INTEL_PT_FUP) {
3257 decoder->set_fup_ptw = true;
3258 no_tip = true;
3259 } else {
3260 intel_pt_log_at("ERROR: Missing FUP after PTWRITE",
3261 decoder->pos);
3262 }
3263 goto next;
3264
3265 case INTEL_PT_PTWRITE:
3266 decoder->state.type = INTEL_PT_PTW;
3267 decoder->state.from_ip = decoder->ip;
3268 decoder->state.to_ip = 0;
3269 decoder->state.ptw_payload = decoder->packet.payload;
3270 return 0;
3271
3272 case INTEL_PT_MWAIT:
3273 decoder->fup_mwait_payload = decoder->packet.payload;
3274 decoder->set_fup_mwait = true;
3275 break;
3276
3277 case INTEL_PT_PWRE:
3278 if (decoder->set_fup_mwait) {
3279 decoder->fup_pwre_payload =
3280 decoder->packet.payload;
3281 decoder->set_fup_pwre = true;
3282 break;
3283 }
3284 decoder->state.type = INTEL_PT_PWR_ENTRY;
3285 decoder->state.from_ip = decoder->ip;
3286 decoder->state.to_ip = 0;
3287 decoder->state.pwrx_payload = decoder->packet.payload;
3288 return 0;
3289
3290 case INTEL_PT_EXSTOP_IP:
3291 err = intel_pt_get_next_packet(decoder);
3292 if (err)
3293 return err;
3294 if (decoder->packet.type == INTEL_PT_FUP) {
3295 decoder->set_fup_exstop = true;
3296 no_tip = true;
3297 } else {
3298 intel_pt_log_at("ERROR: Missing FUP after EXSTOP",
3299 decoder->pos);
3300 }
3301 goto next;
3302
3303 case INTEL_PT_EXSTOP:
3304 decoder->state.type = INTEL_PT_EX_STOP;
3305 decoder->state.from_ip = decoder->ip;
3306 decoder->state.to_ip = 0;
3307 return 0;
3308
3309 case INTEL_PT_PWRX:
3310 decoder->state.type = INTEL_PT_PWR_EXIT;
3311 decoder->state.from_ip = decoder->ip;
3312 decoder->state.to_ip = 0;
3313 decoder->state.pwrx_payload = decoder->packet.payload;
3314 return 0;
3315
3316 case INTEL_PT_BBP:
3317 intel_pt_bbp(decoder);
3318 break;
3319
3320 case INTEL_PT_BIP:
3321 intel_pt_bip(decoder);
3322 break;
3323
3324 case INTEL_PT_BEP:
3325 decoder->state.type = INTEL_PT_BLK_ITEMS;
3326 decoder->state.from_ip = decoder->ip;
3327 decoder->state.to_ip = 0;
3328 return 0;
3329
3330 case INTEL_PT_BEP_IP:
3331 err = intel_pt_get_next_packet(decoder);
3332 if (err)
3333 return err;
3334 if (decoder->packet.type == INTEL_PT_FUP) {
3335 decoder->set_fup_bep = true;
3336 no_tip = true;
3337 } else {
3338 intel_pt_log_at("ERROR: Missing FUP after BEP",
3339 decoder->pos);
3340 }
3341 goto next;
3342
3343 case INTEL_PT_CFE:
3344 case INTEL_PT_CFE_IP:
3345 case INTEL_PT_EVD:
3346 break;
3347
3348 default:
3349 return intel_pt_bug(decoder);
3350 }
3351 }
3352 }
3353
intel_pt_have_ip(struct intel_pt_decoder * decoder)3354 static inline bool intel_pt_have_ip(struct intel_pt_decoder *decoder)
3355 {
3356 return decoder->packet.count &&
3357 (decoder->have_last_ip || decoder->packet.count == 3 ||
3358 decoder->packet.count == 6);
3359 }
3360
3361 /* Walk PSB+ packets to get in sync. */
intel_pt_walk_psb(struct intel_pt_decoder * decoder)3362 static int intel_pt_walk_psb(struct intel_pt_decoder *decoder)
3363 {
3364 int err;
3365
3366 decoder->in_psb = true;
3367
3368 while (1) {
3369 err = intel_pt_get_next_packet(decoder);
3370 if (err)
3371 goto out;
3372
3373 switch (decoder->packet.type) {
3374 case INTEL_PT_TIP_PGD:
3375 decoder->continuous_period = false;
3376 __fallthrough;
3377 case INTEL_PT_TIP_PGE:
3378 case INTEL_PT_TIP:
3379 case INTEL_PT_PTWRITE:
3380 case INTEL_PT_PTWRITE_IP:
3381 case INTEL_PT_EXSTOP:
3382 case INTEL_PT_EXSTOP_IP:
3383 case INTEL_PT_MWAIT:
3384 case INTEL_PT_PWRE:
3385 case INTEL_PT_PWRX:
3386 case INTEL_PT_BBP:
3387 case INTEL_PT_BIP:
3388 case INTEL_PT_BEP:
3389 case INTEL_PT_BEP_IP:
3390 case INTEL_PT_CFE:
3391 case INTEL_PT_CFE_IP:
3392 case INTEL_PT_EVD:
3393 intel_pt_log("ERROR: Unexpected packet\n");
3394 err = -ENOENT;
3395 goto out;
3396
3397 case INTEL_PT_FUP:
3398 decoder->pge = true;
3399 if (intel_pt_have_ip(decoder)) {
3400 uint64_t current_ip = decoder->ip;
3401
3402 intel_pt_set_ip(decoder);
3403 decoder->psb_ip = decoder->ip;
3404 if (current_ip)
3405 intel_pt_log_to("Setting IP",
3406 decoder->ip);
3407 }
3408 break;
3409
3410 case INTEL_PT_MTC:
3411 intel_pt_calc_mtc_timestamp(decoder);
3412 break;
3413
3414 case INTEL_PT_TSC:
3415 intel_pt_calc_tsc_timestamp(decoder);
3416 break;
3417
3418 case INTEL_PT_TMA:
3419 intel_pt_calc_tma(decoder);
3420 break;
3421
3422 case INTEL_PT_CYC:
3423 intel_pt_calc_cyc_timestamp(decoder);
3424 break;
3425
3426 case INTEL_PT_CBR:
3427 intel_pt_calc_cbr(decoder);
3428 break;
3429
3430 case INTEL_PT_PIP:
3431 intel_pt_set_pip(decoder);
3432 break;
3433
3434 case INTEL_PT_MODE_EXEC:
3435 decoder->exec_mode = decoder->packet.payload;
3436 break;
3437
3438 case INTEL_PT_MODE_TSX:
3439 intel_pt_update_in_tx(decoder);
3440 break;
3441
3442 case INTEL_PT_TRACESTOP:
3443 decoder->pge = false;
3444 decoder->continuous_period = false;
3445 intel_pt_clear_tx_flags(decoder);
3446 __fallthrough;
3447
3448 case INTEL_PT_TNT:
3449 decoder->have_tma = false;
3450 intel_pt_log("ERROR: Unexpected packet\n");
3451 if (decoder->ip)
3452 decoder->pkt_state = INTEL_PT_STATE_ERR4;
3453 else
3454 decoder->pkt_state = INTEL_PT_STATE_ERR3;
3455 err = -ENOENT;
3456 goto out;
3457
3458 case INTEL_PT_BAD: /* Does not happen */
3459 err = intel_pt_bug(decoder);
3460 goto out;
3461
3462 case INTEL_PT_OVF:
3463 err = intel_pt_overflow(decoder);
3464 goto out;
3465
3466 case INTEL_PT_PSBEND:
3467 err = 0;
3468 goto out;
3469
3470 case INTEL_PT_PSB:
3471 case INTEL_PT_VMCS:
3472 case INTEL_PT_MNT:
3473 case INTEL_PT_PAD:
3474 default:
3475 break;
3476 }
3477 }
3478 out:
3479 decoder->in_psb = false;
3480
3481 return err;
3482 }
3483
intel_pt_walk_to_ip(struct intel_pt_decoder * decoder)3484 static int intel_pt_walk_to_ip(struct intel_pt_decoder *decoder)
3485 {
3486 int err;
3487
3488 while (1) {
3489 err = intel_pt_get_next_packet(decoder);
3490 if (err)
3491 return err;
3492
3493 switch (decoder->packet.type) {
3494 case INTEL_PT_TIP_PGD:
3495 decoder->continuous_period = false;
3496 decoder->pge = false;
3497 if (intel_pt_have_ip(decoder))
3498 intel_pt_set_ip(decoder);
3499 if (!decoder->ip)
3500 break;
3501 decoder->state.type |= INTEL_PT_TRACE_END;
3502 return 0;
3503
3504 case INTEL_PT_TIP_PGE:
3505 decoder->pge = true;
3506 intel_pt_mtc_cyc_cnt_pge(decoder);
3507 if (intel_pt_have_ip(decoder))
3508 intel_pt_set_ip(decoder);
3509 if (!decoder->ip)
3510 break;
3511 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
3512 return 0;
3513
3514 case INTEL_PT_TIP:
3515 decoder->pge = true;
3516 if (intel_pt_have_ip(decoder))
3517 intel_pt_set_ip(decoder);
3518 if (!decoder->ip)
3519 break;
3520 return 0;
3521
3522 case INTEL_PT_FUP:
3523 if (intel_pt_have_ip(decoder))
3524 intel_pt_set_ip(decoder);
3525 if (decoder->ip)
3526 return 0;
3527 break;
3528
3529 case INTEL_PT_MTC:
3530 intel_pt_calc_mtc_timestamp(decoder);
3531 break;
3532
3533 case INTEL_PT_TSC:
3534 intel_pt_calc_tsc_timestamp(decoder);
3535 break;
3536
3537 case INTEL_PT_TMA:
3538 intel_pt_calc_tma(decoder);
3539 break;
3540
3541 case INTEL_PT_CYC:
3542 intel_pt_calc_cyc_timestamp(decoder);
3543 break;
3544
3545 case INTEL_PT_CBR:
3546 intel_pt_calc_cbr(decoder);
3547 break;
3548
3549 case INTEL_PT_PIP:
3550 intel_pt_set_pip(decoder);
3551 break;
3552
3553 case INTEL_PT_MODE_EXEC:
3554 decoder->exec_mode = decoder->packet.payload;
3555 break;
3556
3557 case INTEL_PT_MODE_TSX:
3558 intel_pt_update_in_tx(decoder);
3559 break;
3560
3561 case INTEL_PT_OVF:
3562 return intel_pt_overflow(decoder);
3563
3564 case INTEL_PT_BAD: /* Does not happen */
3565 return intel_pt_bug(decoder);
3566
3567 case INTEL_PT_TRACESTOP:
3568 decoder->pge = false;
3569 decoder->continuous_period = false;
3570 intel_pt_clear_tx_flags(decoder);
3571 decoder->have_tma = false;
3572 break;
3573
3574 case INTEL_PT_PSB:
3575 decoder->state.psb_offset = decoder->pos;
3576 decoder->psb_ip = 0;
3577 decoder->last_ip = 0;
3578 decoder->have_last_ip = true;
3579 intel_pt_clear_stack(&decoder->stack);
3580 err = intel_pt_walk_psb(decoder);
3581 if (err)
3582 return err;
3583 decoder->state.type = INTEL_PT_PSB_EVT;
3584 decoder->state.from_ip = decoder->psb_ip;
3585 decoder->state.to_ip = 0;
3586 return 0;
3587
3588 case INTEL_PT_TNT:
3589 case INTEL_PT_PSBEND:
3590 case INTEL_PT_VMCS:
3591 case INTEL_PT_MNT:
3592 case INTEL_PT_PAD:
3593 case INTEL_PT_PTWRITE:
3594 case INTEL_PT_PTWRITE_IP:
3595 case INTEL_PT_EXSTOP:
3596 case INTEL_PT_EXSTOP_IP:
3597 case INTEL_PT_MWAIT:
3598 case INTEL_PT_PWRE:
3599 case INTEL_PT_PWRX:
3600 case INTEL_PT_BBP:
3601 case INTEL_PT_BIP:
3602 case INTEL_PT_BEP:
3603 case INTEL_PT_BEP_IP:
3604 case INTEL_PT_CFE:
3605 case INTEL_PT_CFE_IP:
3606 case INTEL_PT_EVD:
3607 default:
3608 break;
3609 }
3610 }
3611 }
3612
intel_pt_sync_ip(struct intel_pt_decoder * decoder)3613 static int intel_pt_sync_ip(struct intel_pt_decoder *decoder)
3614 {
3615 int err;
3616
3617 decoder->set_fup_tx_flags = false;
3618 decoder->set_fup_ptw = false;
3619 decoder->set_fup_mwait = false;
3620 decoder->set_fup_pwre = false;
3621 decoder->set_fup_exstop = false;
3622 decoder->set_fup_bep = false;
3623 decoder->overflow = false;
3624
3625 if (!decoder->branch_enable) {
3626 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3627 decoder->state.type = 0; /* Do not have a sample */
3628 return 0;
3629 }
3630
3631 intel_pt_log("Scanning for full IP\n");
3632 err = intel_pt_walk_to_ip(decoder);
3633 if (err || ((decoder->state.type & INTEL_PT_PSB_EVT) && !decoder->ip))
3634 return err;
3635
3636 /* In hop mode, resample to get the to_ip as an "instruction" sample */
3637 if (decoder->hop)
3638 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
3639 else
3640 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3641
3642 decoder->state.from_ip = 0;
3643 decoder->state.to_ip = decoder->ip;
3644 intel_pt_log_to("Setting IP", decoder->ip);
3645
3646 return 0;
3647 }
3648
intel_pt_part_psb(struct intel_pt_decoder * decoder)3649 static int intel_pt_part_psb(struct intel_pt_decoder *decoder)
3650 {
3651 const unsigned char *end = decoder->buf + decoder->len;
3652 size_t i;
3653
3654 for (i = INTEL_PT_PSB_LEN - 1; i; i--) {
3655 if (i > decoder->len)
3656 continue;
3657 if (!memcmp(end - i, INTEL_PT_PSB_STR, i))
3658 return i;
3659 }
3660 return 0;
3661 }
3662
intel_pt_rest_psb(struct intel_pt_decoder * decoder,int part_psb)3663 static int intel_pt_rest_psb(struct intel_pt_decoder *decoder, int part_psb)
3664 {
3665 size_t rest_psb = INTEL_PT_PSB_LEN - part_psb;
3666 const char *psb = INTEL_PT_PSB_STR;
3667
3668 if (rest_psb > decoder->len ||
3669 memcmp(decoder->buf, psb + part_psb, rest_psb))
3670 return 0;
3671
3672 return rest_psb;
3673 }
3674
intel_pt_get_split_psb(struct intel_pt_decoder * decoder,int part_psb)3675 static int intel_pt_get_split_psb(struct intel_pt_decoder *decoder,
3676 int part_psb)
3677 {
3678 int rest_psb, ret;
3679
3680 decoder->pos += decoder->len;
3681 decoder->len = 0;
3682
3683 ret = intel_pt_get_next_data(decoder, false);
3684 if (ret)
3685 return ret;
3686
3687 rest_psb = intel_pt_rest_psb(decoder, part_psb);
3688 if (!rest_psb)
3689 return 0;
3690
3691 decoder->pos -= part_psb;
3692 decoder->next_buf = decoder->buf + rest_psb;
3693 decoder->next_len = decoder->len - rest_psb;
3694 memcpy(decoder->temp_buf, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
3695 decoder->buf = decoder->temp_buf;
3696 decoder->len = INTEL_PT_PSB_LEN;
3697
3698 return 0;
3699 }
3700
intel_pt_scan_for_psb(struct intel_pt_decoder * decoder)3701 static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder)
3702 {
3703 unsigned char *next;
3704 int ret;
3705
3706 intel_pt_log("Scanning for PSB\n");
3707 while (1) {
3708 if (!decoder->len) {
3709 ret = intel_pt_get_next_data(decoder, false);
3710 if (ret)
3711 return ret;
3712 }
3713
3714 next = memmem(decoder->buf, decoder->len, INTEL_PT_PSB_STR,
3715 INTEL_PT_PSB_LEN);
3716 if (!next) {
3717 int part_psb;
3718
3719 part_psb = intel_pt_part_psb(decoder);
3720 if (part_psb) {
3721 ret = intel_pt_get_split_psb(decoder, part_psb);
3722 if (ret)
3723 return ret;
3724 } else {
3725 decoder->pos += decoder->len;
3726 decoder->len = 0;
3727 }
3728 continue;
3729 }
3730
3731 decoder->pkt_step = next - decoder->buf;
3732 return intel_pt_get_next_packet(decoder);
3733 }
3734 }
3735
intel_pt_sync(struct intel_pt_decoder * decoder)3736 static int intel_pt_sync(struct intel_pt_decoder *decoder)
3737 {
3738 int err;
3739
3740 decoder->pge = false;
3741 decoder->continuous_period = false;
3742 decoder->have_last_ip = false;
3743 decoder->last_ip = 0;
3744 decoder->psb_ip = 0;
3745 decoder->ip = 0;
3746 intel_pt_clear_stack(&decoder->stack);
3747
3748 err = intel_pt_scan_for_psb(decoder);
3749 if (err)
3750 return err;
3751
3752 if (decoder->vm_time_correlation) {
3753 decoder->in_psb = true;
3754 if (!decoder->timestamp)
3755 decoder->timestamp = 1;
3756 decoder->state.type = 0;
3757 decoder->pkt_state = INTEL_PT_STATE_VM_TIME_CORRELATION;
3758 return 0;
3759 }
3760
3761 decoder->have_last_ip = true;
3762 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3763
3764 err = intel_pt_walk_psb(decoder);
3765 if (err)
3766 return err;
3767
3768 decoder->state.type = INTEL_PT_PSB_EVT; /* Only PSB sample */
3769 decoder->state.from_ip = decoder->psb_ip;
3770 decoder->state.to_ip = 0;
3771
3772 if (decoder->ip) {
3773 /*
3774 * In hop mode, resample to get the PSB FUP ip as an
3775 * "instruction" sample.
3776 */
3777 if (decoder->hop)
3778 decoder->pkt_state = INTEL_PT_STATE_RESAMPLE;
3779 else
3780 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
3781 }
3782
3783 return 0;
3784 }
3785
intel_pt_est_timestamp(struct intel_pt_decoder * decoder)3786 static uint64_t intel_pt_est_timestamp(struct intel_pt_decoder *decoder)
3787 {
3788 uint64_t est = decoder->sample_insn_cnt << 1;
3789
3790 if (!decoder->cbr || !decoder->max_non_turbo_ratio)
3791 goto out;
3792
3793 est *= decoder->max_non_turbo_ratio;
3794 est /= decoder->cbr;
3795 out:
3796 return decoder->sample_timestamp + est;
3797 }
3798
intel_pt_decode(struct intel_pt_decoder * decoder)3799 const struct intel_pt_state *intel_pt_decode(struct intel_pt_decoder *decoder)
3800 {
3801 int err;
3802
3803 do {
3804 decoder->state.type = INTEL_PT_BRANCH;
3805 decoder->state.flags = 0;
3806
3807 switch (decoder->pkt_state) {
3808 case INTEL_PT_STATE_NO_PSB:
3809 err = intel_pt_sync(decoder);
3810 break;
3811 case INTEL_PT_STATE_NO_IP:
3812 decoder->have_last_ip = false;
3813 decoder->last_ip = 0;
3814 decoder->ip = 0;
3815 __fallthrough;
3816 case INTEL_PT_STATE_ERR_RESYNC:
3817 err = intel_pt_sync_ip(decoder);
3818 break;
3819 case INTEL_PT_STATE_IN_SYNC:
3820 err = intel_pt_walk_trace(decoder);
3821 break;
3822 case INTEL_PT_STATE_TNT:
3823 case INTEL_PT_STATE_TNT_CONT:
3824 err = intel_pt_walk_tnt(decoder);
3825 if (err == -EAGAIN)
3826 err = intel_pt_walk_trace(decoder);
3827 break;
3828 case INTEL_PT_STATE_TIP:
3829 case INTEL_PT_STATE_TIP_PGD:
3830 err = intel_pt_walk_tip(decoder);
3831 break;
3832 case INTEL_PT_STATE_FUP:
3833 err = intel_pt_walk_fup(decoder);
3834 if (err == -EAGAIN)
3835 err = intel_pt_walk_fup_tip(decoder);
3836 break;
3837 case INTEL_PT_STATE_FUP_NO_TIP:
3838 err = intel_pt_walk_fup(decoder);
3839 if (err == -EAGAIN)
3840 err = intel_pt_walk_trace(decoder);
3841 break;
3842 case INTEL_PT_STATE_FUP_IN_PSB:
3843 err = intel_pt_fup_in_psb(decoder);
3844 break;
3845 case INTEL_PT_STATE_RESAMPLE:
3846 err = intel_pt_resample(decoder);
3847 break;
3848 case INTEL_PT_STATE_VM_TIME_CORRELATION:
3849 err = intel_pt_vm_time_correlation(decoder);
3850 break;
3851 default:
3852 err = intel_pt_bug(decoder);
3853 break;
3854 }
3855 } while (err == -ENOLINK);
3856
3857 if (err) {
3858 decoder->state.err = intel_pt_ext_err(err);
3859 if (err != -EOVERFLOW)
3860 decoder->state.from_ip = decoder->ip;
3861 intel_pt_update_sample_time(decoder);
3862 decoder->sample_tot_cyc_cnt = decoder->tot_cyc_cnt;
3863 intel_pt_set_nr(decoder);
3864 } else {
3865 decoder->state.err = 0;
3866 if (decoder->cbr != decoder->cbr_seen) {
3867 decoder->cbr_seen = decoder->cbr;
3868 if (!decoder->state.type) {
3869 decoder->state.from_ip = decoder->ip;
3870 decoder->state.to_ip = 0;
3871 }
3872 decoder->state.type |= INTEL_PT_CBR_CHG;
3873 decoder->state.cbr_payload = decoder->cbr_payload;
3874 decoder->state.cbr = decoder->cbr;
3875 }
3876 if (intel_pt_sample_time(decoder->pkt_state)) {
3877 intel_pt_update_sample_time(decoder);
3878 if (decoder->sample_cyc) {
3879 decoder->sample_tot_cyc_cnt = decoder->tot_cyc_cnt;
3880 decoder->state.flags |= INTEL_PT_SAMPLE_IPC;
3881 decoder->sample_cyc = false;
3882 }
3883 }
3884 /*
3885 * When using only TSC/MTC to compute cycles, IPC can be
3886 * sampled as soon as the cycle count changes.
3887 */
3888 if (!decoder->have_cyc)
3889 decoder->state.flags |= INTEL_PT_SAMPLE_IPC;
3890 }
3891
3892 /* Let PSB event always have TSC timestamp */
3893 if ((decoder->state.type & INTEL_PT_PSB_EVT) && decoder->tsc_timestamp)
3894 decoder->sample_timestamp = decoder->tsc_timestamp;
3895
3896 decoder->state.from_nr = decoder->nr;
3897 decoder->state.to_nr = decoder->next_nr;
3898 decoder->nr = decoder->next_nr;
3899
3900 decoder->state.timestamp = decoder->sample_timestamp;
3901 decoder->state.est_timestamp = intel_pt_est_timestamp(decoder);
3902 decoder->state.tot_insn_cnt = decoder->tot_insn_cnt;
3903 decoder->state.tot_cyc_cnt = decoder->sample_tot_cyc_cnt;
3904
3905 return &decoder->state;
3906 }
3907
3908 /**
3909 * intel_pt_next_psb - move buffer pointer to the start of the next PSB packet.
3910 * @buf: pointer to buffer pointer
3911 * @len: size of buffer
3912 *
3913 * Updates the buffer pointer to point to the start of the next PSB packet if
3914 * there is one, otherwise the buffer pointer is unchanged. If @buf is updated,
3915 * @len is adjusted accordingly.
3916 *
3917 * Return: %true if a PSB packet is found, %false otherwise.
3918 */
intel_pt_next_psb(unsigned char ** buf,size_t * len)3919 static bool intel_pt_next_psb(unsigned char **buf, size_t *len)
3920 {
3921 unsigned char *next;
3922
3923 next = memmem(*buf, *len, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
3924 if (next) {
3925 *len -= next - *buf;
3926 *buf = next;
3927 return true;
3928 }
3929 return false;
3930 }
3931
3932 /**
3933 * intel_pt_step_psb - move buffer pointer to the start of the following PSB
3934 * packet.
3935 * @buf: pointer to buffer pointer
3936 * @len: size of buffer
3937 *
3938 * Updates the buffer pointer to point to the start of the following PSB packet
3939 * (skipping the PSB at @buf itself) if there is one, otherwise the buffer
3940 * pointer is unchanged. If @buf is updated, @len is adjusted accordingly.
3941 *
3942 * Return: %true if a PSB packet is found, %false otherwise.
3943 */
intel_pt_step_psb(unsigned char ** buf,size_t * len)3944 static bool intel_pt_step_psb(unsigned char **buf, size_t *len)
3945 {
3946 unsigned char *next;
3947
3948 if (!*len)
3949 return false;
3950
3951 next = memmem(*buf + 1, *len - 1, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
3952 if (next) {
3953 *len -= next - *buf;
3954 *buf = next;
3955 return true;
3956 }
3957 return false;
3958 }
3959
3960 /**
3961 * intel_pt_last_psb - find the last PSB packet in a buffer.
3962 * @buf: buffer
3963 * @len: size of buffer
3964 *
3965 * This function finds the last PSB in a buffer.
3966 *
3967 * Return: A pointer to the last PSB in @buf if found, %NULL otherwise.
3968 */
intel_pt_last_psb(unsigned char * buf,size_t len)3969 static unsigned char *intel_pt_last_psb(unsigned char *buf, size_t len)
3970 {
3971 const char *n = INTEL_PT_PSB_STR;
3972 unsigned char *p;
3973 size_t k;
3974
3975 if (len < INTEL_PT_PSB_LEN)
3976 return NULL;
3977
3978 k = len - INTEL_PT_PSB_LEN + 1;
3979 while (1) {
3980 p = memrchr(buf, n[0], k);
3981 if (!p)
3982 return NULL;
3983 if (!memcmp(p + 1, n + 1, INTEL_PT_PSB_LEN - 1))
3984 return p;
3985 k = p - buf;
3986 if (!k)
3987 return NULL;
3988 }
3989 }
3990
3991 /**
3992 * intel_pt_next_tsc - find and return next TSC.
3993 * @buf: buffer
3994 * @len: size of buffer
3995 * @tsc: TSC value returned
3996 * @rem: returns remaining size when TSC is found
3997 *
3998 * Find a TSC packet in @buf and return the TSC value. This function assumes
3999 * that @buf starts at a PSB and that PSB+ will contain TSC and so stops if a
4000 * PSBEND packet is found.
4001 *
4002 * Return: %true if TSC is found, false otherwise.
4003 */
intel_pt_next_tsc(unsigned char * buf,size_t len,uint64_t * tsc,size_t * rem)4004 static bool intel_pt_next_tsc(unsigned char *buf, size_t len, uint64_t *tsc,
4005 size_t *rem)
4006 {
4007 enum intel_pt_pkt_ctx ctx = INTEL_PT_NO_CTX;
4008 struct intel_pt_pkt packet;
4009 int ret;
4010
4011 while (len) {
4012 ret = intel_pt_get_packet(buf, len, &packet, &ctx);
4013 if (ret <= 0)
4014 return false;
4015 if (packet.type == INTEL_PT_TSC) {
4016 *tsc = packet.payload;
4017 *rem = len;
4018 return true;
4019 }
4020 if (packet.type == INTEL_PT_PSBEND)
4021 return false;
4022 buf += ret;
4023 len -= ret;
4024 }
4025 return false;
4026 }
4027
4028 /**
4029 * intel_pt_tsc_cmp - compare 7-byte TSCs.
4030 * @tsc1: first TSC to compare
4031 * @tsc2: second TSC to compare
4032 *
4033 * This function compares 7-byte TSC values allowing for the possibility that
4034 * TSC wrapped around. Generally it is not possible to know if TSC has wrapped
4035 * around so for that purpose this function assumes the absolute difference is
4036 * less than half the maximum difference.
4037 *
4038 * Return: %-1 if @tsc1 is before @tsc2, %0 if @tsc1 == @tsc2, %1 if @tsc1 is
4039 * after @tsc2.
4040 */
intel_pt_tsc_cmp(uint64_t tsc1,uint64_t tsc2)4041 static int intel_pt_tsc_cmp(uint64_t tsc1, uint64_t tsc2)
4042 {
4043 const uint64_t halfway = (1ULL << 55);
4044
4045 if (tsc1 == tsc2)
4046 return 0;
4047
4048 if (tsc1 < tsc2) {
4049 if (tsc2 - tsc1 < halfway)
4050 return -1;
4051 else
4052 return 1;
4053 } else {
4054 if (tsc1 - tsc2 < halfway)
4055 return 1;
4056 else
4057 return -1;
4058 }
4059 }
4060
4061 #define MAX_PADDING (PERF_AUXTRACE_RECORD_ALIGNMENT - 1)
4062
4063 /**
4064 * adj_for_padding - adjust overlap to account for padding.
4065 * @buf_b: second buffer
4066 * @buf_a: first buffer
4067 * @len_a: size of first buffer
4068 *
4069 * @buf_a might have up to 7 bytes of padding appended. Adjust the overlap
4070 * accordingly.
4071 *
4072 * Return: A pointer into @buf_b from where non-overlapped data starts
4073 */
adj_for_padding(unsigned char * buf_b,unsigned char * buf_a,size_t len_a)4074 static unsigned char *adj_for_padding(unsigned char *buf_b,
4075 unsigned char *buf_a, size_t len_a)
4076 {
4077 unsigned char *p = buf_b - MAX_PADDING;
4078 unsigned char *q = buf_a + len_a - MAX_PADDING;
4079 int i;
4080
4081 for (i = MAX_PADDING; i; i--, p++, q++) {
4082 if (*p != *q)
4083 break;
4084 }
4085
4086 return p;
4087 }
4088
4089 /**
4090 * intel_pt_find_overlap_tsc - determine start of non-overlapped trace data
4091 * using TSC.
4092 * @buf_a: first buffer
4093 * @len_a: size of first buffer
4094 * @buf_b: second buffer
4095 * @len_b: size of second buffer
4096 * @consecutive: returns true if there is data in buf_b that is consecutive
4097 * to buf_a
4098 * @ooo_tsc: out-of-order TSC due to VM TSC offset / scaling
4099 *
4100 * If the trace contains TSC we can look at the last TSC of @buf_a and the
4101 * first TSC of @buf_b in order to determine if the buffers overlap, and then
4102 * walk forward in @buf_b until a later TSC is found. A precondition is that
4103 * @buf_a and @buf_b are positioned at a PSB.
4104 *
4105 * Return: A pointer into @buf_b from where non-overlapped data starts, or
4106 * @buf_b + @len_b if there is no non-overlapped data.
4107 */
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)4108 static unsigned char *intel_pt_find_overlap_tsc(unsigned char *buf_a,
4109 size_t len_a,
4110 unsigned char *buf_b,
4111 size_t len_b, bool *consecutive,
4112 bool ooo_tsc)
4113 {
4114 uint64_t tsc_a, tsc_b;
4115 unsigned char *p;
4116 size_t len, rem_a, rem_b;
4117
4118 p = intel_pt_last_psb(buf_a, len_a);
4119 if (!p)
4120 return buf_b; /* No PSB in buf_a => no overlap */
4121
4122 len = len_a - (p - buf_a);
4123 if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a)) {
4124 /* The last PSB+ in buf_a is incomplete, so go back one more */
4125 len_a -= len;
4126 p = intel_pt_last_psb(buf_a, len_a);
4127 if (!p)
4128 return buf_b; /* No full PSB+ => assume no overlap */
4129 len = len_a - (p - buf_a);
4130 if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a))
4131 return buf_b; /* No TSC in buf_a => assume no overlap */
4132 }
4133
4134 while (1) {
4135 /* Ignore PSB+ with no TSC */
4136 if (intel_pt_next_tsc(buf_b, len_b, &tsc_b, &rem_b)) {
4137 int cmp = intel_pt_tsc_cmp(tsc_a, tsc_b);
4138
4139 /* Same TSC, so buffers are consecutive */
4140 if (!cmp && rem_b >= rem_a) {
4141 unsigned char *start;
4142
4143 *consecutive = true;
4144 start = buf_b + len_b - (rem_b - rem_a);
4145 return adj_for_padding(start, buf_a, len_a);
4146 }
4147 if (cmp < 0 && !ooo_tsc)
4148 return buf_b; /* tsc_a < tsc_b => no overlap */
4149 }
4150
4151 if (!intel_pt_step_psb(&buf_b, &len_b))
4152 return buf_b + len_b; /* No PSB in buf_b => no data */
4153 }
4154 }
4155
4156 /**
4157 * intel_pt_find_overlap - determine start of non-overlapped trace data.
4158 * @buf_a: first buffer
4159 * @len_a: size of first buffer
4160 * @buf_b: second buffer
4161 * @len_b: size of second buffer
4162 * @have_tsc: can use TSC packets to detect overlap
4163 * @consecutive: returns true if there is data in buf_b that is consecutive
4164 * to buf_a
4165 * @ooo_tsc: out-of-order TSC due to VM TSC offset / scaling
4166 *
4167 * When trace samples or snapshots are recorded there is the possibility that
4168 * the data overlaps. Note that, for the purposes of decoding, data is only
4169 * useful if it begins with a PSB packet.
4170 *
4171 * Return: A pointer into @buf_b from where non-overlapped data starts, or
4172 * @buf_b + @len_b if there is no non-overlapped data.
4173 */
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)4174 unsigned char *intel_pt_find_overlap(unsigned char *buf_a, size_t len_a,
4175 unsigned char *buf_b, size_t len_b,
4176 bool have_tsc, bool *consecutive,
4177 bool ooo_tsc)
4178 {
4179 unsigned char *found;
4180
4181 /* Buffer 'b' must start at PSB so throw away everything before that */
4182 if (!intel_pt_next_psb(&buf_b, &len_b))
4183 return buf_b + len_b; /* No PSB */
4184
4185 if (!intel_pt_next_psb(&buf_a, &len_a))
4186 return buf_b; /* No overlap */
4187
4188 if (have_tsc) {
4189 found = intel_pt_find_overlap_tsc(buf_a, len_a, buf_b, len_b,
4190 consecutive, ooo_tsc);
4191 if (found)
4192 return found;
4193 }
4194
4195 /*
4196 * Buffer 'b' cannot end within buffer 'a' so, for comparison purposes,
4197 * we can ignore the first part of buffer 'a'.
4198 */
4199 while (len_b < len_a) {
4200 if (!intel_pt_step_psb(&buf_a, &len_a))
4201 return buf_b; /* No overlap */
4202 }
4203
4204 /* Now len_b >= len_a */
4205 while (1) {
4206 /* Potential overlap so check the bytes */
4207 found = memmem(buf_a, len_a, buf_b, len_a);
4208 if (found) {
4209 *consecutive = true;
4210 return adj_for_padding(buf_b + len_a, buf_a, len_a);
4211 }
4212
4213 /* Try again at next PSB in buffer 'a' */
4214 if (!intel_pt_step_psb(&buf_a, &len_a))
4215 return buf_b; /* No overlap */
4216 }
4217 }
4218
4219 /**
4220 * struct fast_forward_data - data used by intel_pt_ff_cb().
4221 * @timestamp: timestamp to fast forward towards
4222 * @buf_timestamp: buffer timestamp of last buffer with trace data earlier than
4223 * the fast forward timestamp.
4224 */
4225 struct fast_forward_data {
4226 uint64_t timestamp;
4227 uint64_t buf_timestamp;
4228 };
4229
4230 /**
4231 * intel_pt_ff_cb - fast forward lookahead callback.
4232 * @buffer: Intel PT trace buffer
4233 * @data: opaque pointer to fast forward data (struct fast_forward_data)
4234 *
4235 * Determine if @buffer trace is past the fast forward timestamp.
4236 *
4237 * Return: 1 (stop lookahead) if @buffer trace is past the fast forward
4238 * timestamp, and 0 otherwise.
4239 */
intel_pt_ff_cb(struct intel_pt_buffer * buffer,void * data)4240 static int intel_pt_ff_cb(struct intel_pt_buffer *buffer, void *data)
4241 {
4242 struct fast_forward_data *d = data;
4243 unsigned char *buf;
4244 uint64_t tsc;
4245 size_t rem;
4246 size_t len;
4247
4248 buf = (unsigned char *)buffer->buf;
4249 len = buffer->len;
4250
4251 if (!intel_pt_next_psb(&buf, &len) ||
4252 !intel_pt_next_tsc(buf, len, &tsc, &rem))
4253 return 0;
4254
4255 tsc = intel_pt_8b_tsc(tsc, buffer->ref_timestamp);
4256
4257 intel_pt_log("Buffer 1st timestamp " x64_fmt " ref timestamp " x64_fmt "\n",
4258 tsc, buffer->ref_timestamp);
4259
4260 /*
4261 * If the buffer contains a timestamp earlier that the fast forward
4262 * timestamp, then record it, else stop.
4263 */
4264 if (tsc < d->timestamp)
4265 d->buf_timestamp = buffer->ref_timestamp;
4266 else
4267 return 1;
4268
4269 return 0;
4270 }
4271
4272 /**
4273 * intel_pt_fast_forward - reposition decoder forwards.
4274 * @decoder: Intel PT decoder
4275 * @timestamp: timestamp to fast forward towards
4276 *
4277 * Reposition decoder at the last PSB with a timestamp earlier than @timestamp.
4278 *
4279 * Return: 0 on success or negative error code on failure.
4280 */
intel_pt_fast_forward(struct intel_pt_decoder * decoder,uint64_t timestamp)4281 int intel_pt_fast_forward(struct intel_pt_decoder *decoder, uint64_t timestamp)
4282 {
4283 struct fast_forward_data d = { .timestamp = timestamp };
4284 unsigned char *buf;
4285 size_t len;
4286 int err;
4287
4288 intel_pt_log("Fast forward towards timestamp " x64_fmt "\n", timestamp);
4289
4290 /* Find buffer timestamp of buffer to fast forward to */
4291 err = decoder->lookahead(decoder->data, intel_pt_ff_cb, &d);
4292 if (err < 0)
4293 return err;
4294
4295 /* Walk to buffer with same buffer timestamp */
4296 if (d.buf_timestamp) {
4297 do {
4298 decoder->pos += decoder->len;
4299 decoder->len = 0;
4300 err = intel_pt_get_next_data(decoder, true);
4301 /* -ENOLINK means non-consecutive trace */
4302 if (err && err != -ENOLINK)
4303 return err;
4304 } while (decoder->buf_timestamp != d.buf_timestamp);
4305 }
4306
4307 if (!decoder->buf)
4308 return 0;
4309
4310 buf = (unsigned char *)decoder->buf;
4311 len = decoder->len;
4312
4313 if (!intel_pt_next_psb(&buf, &len))
4314 return 0;
4315
4316 /*
4317 * Walk PSBs while the PSB timestamp is less than the fast forward
4318 * timestamp.
4319 */
4320 do {
4321 uint64_t tsc;
4322 size_t rem;
4323
4324 if (!intel_pt_next_tsc(buf, len, &tsc, &rem))
4325 break;
4326 tsc = intel_pt_8b_tsc(tsc, decoder->buf_timestamp);
4327 /*
4328 * A TSC packet can slip past MTC packets but, after fast
4329 * forward, decoding starts at the TSC timestamp. That means
4330 * the timestamps may not be exactly the same as the timestamps
4331 * that would have been decoded without fast forward.
4332 */
4333 if (tsc < timestamp) {
4334 intel_pt_log("Fast forward to next PSB timestamp " x64_fmt "\n", tsc);
4335 decoder->pos += decoder->len - len;
4336 decoder->buf = buf;
4337 decoder->len = len;
4338 intel_pt_reposition(decoder);
4339 } else {
4340 break;
4341 }
4342 } while (intel_pt_step_psb(&buf, &len));
4343
4344 return 0;
4345 }
4346