Merge branch 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[muen/linux.git] / tools / perf / util / intel-pt-decoder / intel-pt-decoder.c
1 /*
2  * intel_pt_decoder.c: Intel Processor Trace support
3  * Copyright (c) 2013-2014, Intel Corporation.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms and conditions of the GNU General Public License,
7  * version 2, as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12  * more details.
13  *
14  */
15
16 #ifndef _GNU_SOURCE
17 #define _GNU_SOURCE
18 #endif
19 #include <stdlib.h>
20 #include <stdbool.h>
21 #include <string.h>
22 #include <errno.h>
23 #include <stdint.h>
24 #include <inttypes.h>
25 #include <linux/compiler.h>
26
27 #include "../cache.h"
28 #include "../util.h"
29
30 #include "intel-pt-insn-decoder.h"
31 #include "intel-pt-pkt-decoder.h"
32 #include "intel-pt-decoder.h"
33 #include "intel-pt-log.h"
34
35 #define INTEL_PT_BLK_SIZE 1024
36
37 #define BIT63 (((uint64_t)1 << 63))
38
39 #define INTEL_PT_RETURN 1
40
41 /* Maximum number of loops with no packets consumed i.e. stuck in a loop */
42 #define INTEL_PT_MAX_LOOPS 10000
43
44 struct intel_pt_blk {
45         struct intel_pt_blk *prev;
46         uint64_t ip[INTEL_PT_BLK_SIZE];
47 };
48
49 struct intel_pt_stack {
50         struct intel_pt_blk *blk;
51         struct intel_pt_blk *spare;
52         int pos;
53 };
54
55 enum intel_pt_pkt_state {
56         INTEL_PT_STATE_NO_PSB,
57         INTEL_PT_STATE_NO_IP,
58         INTEL_PT_STATE_ERR_RESYNC,
59         INTEL_PT_STATE_IN_SYNC,
60         INTEL_PT_STATE_TNT,
61         INTEL_PT_STATE_TIP,
62         INTEL_PT_STATE_TIP_PGD,
63         INTEL_PT_STATE_FUP,
64         INTEL_PT_STATE_FUP_NO_TIP,
65 };
66
67 static inline bool intel_pt_sample_time(enum intel_pt_pkt_state pkt_state)
68 {
69         switch (pkt_state) {
70         case INTEL_PT_STATE_NO_PSB:
71         case INTEL_PT_STATE_NO_IP:
72         case INTEL_PT_STATE_ERR_RESYNC:
73         case INTEL_PT_STATE_IN_SYNC:
74         case INTEL_PT_STATE_TNT:
75                 return true;
76         case INTEL_PT_STATE_TIP:
77         case INTEL_PT_STATE_TIP_PGD:
78         case INTEL_PT_STATE_FUP:
79         case INTEL_PT_STATE_FUP_NO_TIP:
80                 return false;
81         default:
82                 return true;
83         };
84 }
85
86 #ifdef INTEL_PT_STRICT
87 #define INTEL_PT_STATE_ERR1     INTEL_PT_STATE_NO_PSB
88 #define INTEL_PT_STATE_ERR2     INTEL_PT_STATE_NO_PSB
89 #define INTEL_PT_STATE_ERR3     INTEL_PT_STATE_NO_PSB
90 #define INTEL_PT_STATE_ERR4     INTEL_PT_STATE_NO_PSB
91 #else
92 #define INTEL_PT_STATE_ERR1     (decoder->pkt_state)
93 #define INTEL_PT_STATE_ERR2     INTEL_PT_STATE_NO_IP
94 #define INTEL_PT_STATE_ERR3     INTEL_PT_STATE_ERR_RESYNC
95 #define INTEL_PT_STATE_ERR4     INTEL_PT_STATE_IN_SYNC
96 #endif
97
98 struct intel_pt_decoder {
99         int (*get_trace)(struct intel_pt_buffer *buffer, void *data);
100         int (*walk_insn)(struct intel_pt_insn *intel_pt_insn,
101                          uint64_t *insn_cnt_ptr, uint64_t *ip, uint64_t to_ip,
102                          uint64_t max_insn_cnt, void *data);
103         bool (*pgd_ip)(uint64_t ip, void *data);
104         void *data;
105         struct intel_pt_state state;
106         const unsigned char *buf;
107         size_t len;
108         bool return_compression;
109         bool branch_enable;
110         bool mtc_insn;
111         bool pge;
112         bool have_tma;
113         bool have_cyc;
114         bool fixup_last_mtc;
115         bool have_last_ip;
116         enum intel_pt_param_flags flags;
117         uint64_t pos;
118         uint64_t last_ip;
119         uint64_t ip;
120         uint64_t cr3;
121         uint64_t timestamp;
122         uint64_t tsc_timestamp;
123         uint64_t ref_timestamp;
124         uint64_t sample_timestamp;
125         uint64_t ret_addr;
126         uint64_t ctc_timestamp;
127         uint64_t ctc_delta;
128         uint64_t cycle_cnt;
129         uint64_t cyc_ref_timestamp;
130         uint32_t last_mtc;
131         uint32_t tsc_ctc_ratio_n;
132         uint32_t tsc_ctc_ratio_d;
133         uint32_t tsc_ctc_mult;
134         uint32_t tsc_slip;
135         uint32_t ctc_rem_mask;
136         int mtc_shift;
137         struct intel_pt_stack stack;
138         enum intel_pt_pkt_state pkt_state;
139         struct intel_pt_pkt packet;
140         struct intel_pt_pkt tnt;
141         int pkt_step;
142         int pkt_len;
143         int last_packet_type;
144         unsigned int cbr;
145         unsigned int cbr_seen;
146         unsigned int max_non_turbo_ratio;
147         double max_non_turbo_ratio_fp;
148         double cbr_cyc_to_tsc;
149         double calc_cyc_to_tsc;
150         bool have_calc_cyc_to_tsc;
151         int exec_mode;
152         unsigned int insn_bytes;
153         uint64_t period;
154         enum intel_pt_period_type period_type;
155         uint64_t tot_insn_cnt;
156         uint64_t period_insn_cnt;
157         uint64_t period_mask;
158         uint64_t period_ticks;
159         uint64_t last_masked_timestamp;
160         bool continuous_period;
161         bool overflow;
162         bool set_fup_tx_flags;
163         bool set_fup_ptw;
164         bool set_fup_mwait;
165         bool set_fup_pwre;
166         bool set_fup_exstop;
167         unsigned int fup_tx_flags;
168         unsigned int tx_flags;
169         uint64_t fup_ptw_payload;
170         uint64_t fup_mwait_payload;
171         uint64_t fup_pwre_payload;
172         uint64_t cbr_payload;
173         uint64_t timestamp_insn_cnt;
174         uint64_t sample_insn_cnt;
175         uint64_t stuck_ip;
176         int no_progress;
177         int stuck_ip_prd;
178         int stuck_ip_cnt;
179         const unsigned char *next_buf;
180         size_t next_len;
181         unsigned char temp_buf[INTEL_PT_PKT_MAX_SZ];
182 };
183
184 static uint64_t intel_pt_lower_power_of_2(uint64_t x)
185 {
186         int i;
187
188         for (i = 0; x != 1; i++)
189                 x >>= 1;
190
191         return x << i;
192 }
193
194 static void intel_pt_setup_period(struct intel_pt_decoder *decoder)
195 {
196         if (decoder->period_type == INTEL_PT_PERIOD_TICKS) {
197                 uint64_t period;
198
199                 period = intel_pt_lower_power_of_2(decoder->period);
200                 decoder->period_mask  = ~(period - 1);
201                 decoder->period_ticks = period;
202         }
203 }
204
205 static uint64_t multdiv(uint64_t t, uint32_t n, uint32_t d)
206 {
207         if (!d)
208                 return 0;
209         return (t / d) * n + ((t % d) * n) / d;
210 }
211
212 struct intel_pt_decoder *intel_pt_decoder_new(struct intel_pt_params *params)
213 {
214         struct intel_pt_decoder *decoder;
215
216         if (!params->get_trace || !params->walk_insn)
217                 return NULL;
218
219         decoder = zalloc(sizeof(struct intel_pt_decoder));
220         if (!decoder)
221                 return NULL;
222
223         decoder->get_trace          = params->get_trace;
224         decoder->walk_insn          = params->walk_insn;
225         decoder->pgd_ip             = params->pgd_ip;
226         decoder->data               = params->data;
227         decoder->return_compression = params->return_compression;
228         decoder->branch_enable      = params->branch_enable;
229
230         decoder->flags              = params->flags;
231
232         decoder->period             = params->period;
233         decoder->period_type        = params->period_type;
234
235         decoder->max_non_turbo_ratio    = params->max_non_turbo_ratio;
236         decoder->max_non_turbo_ratio_fp = params->max_non_turbo_ratio;
237
238         intel_pt_setup_period(decoder);
239
240         decoder->mtc_shift = params->mtc_period;
241         decoder->ctc_rem_mask = (1 << decoder->mtc_shift) - 1;
242
243         decoder->tsc_ctc_ratio_n = params->tsc_ctc_ratio_n;
244         decoder->tsc_ctc_ratio_d = params->tsc_ctc_ratio_d;
245
246         if (!decoder->tsc_ctc_ratio_n)
247                 decoder->tsc_ctc_ratio_d = 0;
248
249         if (decoder->tsc_ctc_ratio_d) {
250                 if (!(decoder->tsc_ctc_ratio_n % decoder->tsc_ctc_ratio_d))
251                         decoder->tsc_ctc_mult = decoder->tsc_ctc_ratio_n /
252                                                 decoder->tsc_ctc_ratio_d;
253
254                 /*
255                  * Allow for timestamps appearing to backwards because a TSC
256                  * packet has slipped past a MTC packet, so allow 2 MTC ticks
257                  * or ...
258                  */
259                 decoder->tsc_slip = multdiv(2 << decoder->mtc_shift,
260                                         decoder->tsc_ctc_ratio_n,
261                                         decoder->tsc_ctc_ratio_d);
262         }
263         /* ... or 0x100 paranoia */
264         if (decoder->tsc_slip < 0x100)
265                 decoder->tsc_slip = 0x100;
266
267         intel_pt_log("timestamp: mtc_shift %u\n", decoder->mtc_shift);
268         intel_pt_log("timestamp: tsc_ctc_ratio_n %u\n", decoder->tsc_ctc_ratio_n);
269         intel_pt_log("timestamp: tsc_ctc_ratio_d %u\n", decoder->tsc_ctc_ratio_d);
270         intel_pt_log("timestamp: tsc_ctc_mult %u\n", decoder->tsc_ctc_mult);
271         intel_pt_log("timestamp: tsc_slip %#x\n", decoder->tsc_slip);
272
273         return decoder;
274 }
275
276 static void intel_pt_pop_blk(struct intel_pt_stack *stack)
277 {
278         struct intel_pt_blk *blk = stack->blk;
279
280         stack->blk = blk->prev;
281         if (!stack->spare)
282                 stack->spare = blk;
283         else
284                 free(blk);
285 }
286
287 static uint64_t intel_pt_pop(struct intel_pt_stack *stack)
288 {
289         if (!stack->pos) {
290                 if (!stack->blk)
291                         return 0;
292                 intel_pt_pop_blk(stack);
293                 if (!stack->blk)
294                         return 0;
295                 stack->pos = INTEL_PT_BLK_SIZE;
296         }
297         return stack->blk->ip[--stack->pos];
298 }
299
300 static int intel_pt_alloc_blk(struct intel_pt_stack *stack)
301 {
302         struct intel_pt_blk *blk;
303
304         if (stack->spare) {
305                 blk = stack->spare;
306                 stack->spare = NULL;
307         } else {
308                 blk = malloc(sizeof(struct intel_pt_blk));
309                 if (!blk)
310                         return -ENOMEM;
311         }
312
313         blk->prev = stack->blk;
314         stack->blk = blk;
315         stack->pos = 0;
316         return 0;
317 }
318
319 static int intel_pt_push(struct intel_pt_stack *stack, uint64_t ip)
320 {
321         int err;
322
323         if (!stack->blk || stack->pos == INTEL_PT_BLK_SIZE) {
324                 err = intel_pt_alloc_blk(stack);
325                 if (err)
326                         return err;
327         }
328
329         stack->blk->ip[stack->pos++] = ip;
330         return 0;
331 }
332
333 static void intel_pt_clear_stack(struct intel_pt_stack *stack)
334 {
335         while (stack->blk)
336                 intel_pt_pop_blk(stack);
337         stack->pos = 0;
338 }
339
340 static void intel_pt_free_stack(struct intel_pt_stack *stack)
341 {
342         intel_pt_clear_stack(stack);
343         zfree(&stack->blk);
344         zfree(&stack->spare);
345 }
346
347 void intel_pt_decoder_free(struct intel_pt_decoder *decoder)
348 {
349         intel_pt_free_stack(&decoder->stack);
350         free(decoder);
351 }
352
353 static int intel_pt_ext_err(int code)
354 {
355         switch (code) {
356         case -ENOMEM:
357                 return INTEL_PT_ERR_NOMEM;
358         case -ENOSYS:
359                 return INTEL_PT_ERR_INTERN;
360         case -EBADMSG:
361                 return INTEL_PT_ERR_BADPKT;
362         case -ENODATA:
363                 return INTEL_PT_ERR_NODATA;
364         case -EILSEQ:
365                 return INTEL_PT_ERR_NOINSN;
366         case -ENOENT:
367                 return INTEL_PT_ERR_MISMAT;
368         case -EOVERFLOW:
369                 return INTEL_PT_ERR_OVR;
370         case -ENOSPC:
371                 return INTEL_PT_ERR_LOST;
372         case -ELOOP:
373                 return INTEL_PT_ERR_NELOOP;
374         default:
375                 return INTEL_PT_ERR_UNK;
376         }
377 }
378
379 static const char *intel_pt_err_msgs[] = {
380         [INTEL_PT_ERR_NOMEM]  = "Memory allocation failed",
381         [INTEL_PT_ERR_INTERN] = "Internal error",
382         [INTEL_PT_ERR_BADPKT] = "Bad packet",
383         [INTEL_PT_ERR_NODATA] = "No more data",
384         [INTEL_PT_ERR_NOINSN] = "Failed to get instruction",
385         [INTEL_PT_ERR_MISMAT] = "Trace doesn't match instruction",
386         [INTEL_PT_ERR_OVR]    = "Overflow packet",
387         [INTEL_PT_ERR_LOST]   = "Lost trace data",
388         [INTEL_PT_ERR_UNK]    = "Unknown error!",
389         [INTEL_PT_ERR_NELOOP] = "Never-ending loop",
390 };
391
392 int intel_pt__strerror(int code, char *buf, size_t buflen)
393 {
394         if (code < 1 || code >= INTEL_PT_ERR_MAX)
395                 code = INTEL_PT_ERR_UNK;
396         strlcpy(buf, intel_pt_err_msgs[code], buflen);
397         return 0;
398 }
399
400 static uint64_t intel_pt_calc_ip(const struct intel_pt_pkt *packet,
401                                  uint64_t last_ip)
402 {
403         uint64_t ip;
404
405         switch (packet->count) {
406         case 1:
407                 ip = (last_ip & (uint64_t)0xffffffffffff0000ULL) |
408                      packet->payload;
409                 break;
410         case 2:
411                 ip = (last_ip & (uint64_t)0xffffffff00000000ULL) |
412                      packet->payload;
413                 break;
414         case 3:
415                 ip = packet->payload;
416                 /* Sign-extend 6-byte ip */
417                 if (ip & (uint64_t)0x800000000000ULL)
418                         ip |= (uint64_t)0xffff000000000000ULL;
419                 break;
420         case 4:
421                 ip = (last_ip & (uint64_t)0xffff000000000000ULL) |
422                      packet->payload;
423                 break;
424         case 6:
425                 ip = packet->payload;
426                 break;
427         default:
428                 return 0;
429         }
430
431         return ip;
432 }
433
434 static inline void intel_pt_set_last_ip(struct intel_pt_decoder *decoder)
435 {
436         decoder->last_ip = intel_pt_calc_ip(&decoder->packet, decoder->last_ip);
437         decoder->have_last_ip = true;
438 }
439
440 static inline void intel_pt_set_ip(struct intel_pt_decoder *decoder)
441 {
442         intel_pt_set_last_ip(decoder);
443         decoder->ip = decoder->last_ip;
444 }
445
446 static void intel_pt_decoder_log_packet(struct intel_pt_decoder *decoder)
447 {
448         intel_pt_log_packet(&decoder->packet, decoder->pkt_len, decoder->pos,
449                             decoder->buf);
450 }
451
452 static int intel_pt_bug(struct intel_pt_decoder *decoder)
453 {
454         intel_pt_log("ERROR: Internal error\n");
455         decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
456         return -ENOSYS;
457 }
458
459 static inline void intel_pt_clear_tx_flags(struct intel_pt_decoder *decoder)
460 {
461         decoder->tx_flags = 0;
462 }
463
464 static inline void intel_pt_update_in_tx(struct intel_pt_decoder *decoder)
465 {
466         decoder->tx_flags = decoder->packet.payload & INTEL_PT_IN_TX;
467 }
468
469 static int intel_pt_bad_packet(struct intel_pt_decoder *decoder)
470 {
471         intel_pt_clear_tx_flags(decoder);
472         decoder->have_tma = false;
473         decoder->pkt_len = 1;
474         decoder->pkt_step = 1;
475         intel_pt_decoder_log_packet(decoder);
476         if (decoder->pkt_state != INTEL_PT_STATE_NO_PSB) {
477                 intel_pt_log("ERROR: Bad packet\n");
478                 decoder->pkt_state = INTEL_PT_STATE_ERR1;
479         }
480         return -EBADMSG;
481 }
482
483 static int intel_pt_get_data(struct intel_pt_decoder *decoder)
484 {
485         struct intel_pt_buffer buffer = { .buf = 0, };
486         int ret;
487
488         decoder->pkt_step = 0;
489
490         intel_pt_log("Getting more data\n");
491         ret = decoder->get_trace(&buffer, decoder->data);
492         if (ret)
493                 return ret;
494         decoder->buf = buffer.buf;
495         decoder->len = buffer.len;
496         if (!decoder->len) {
497                 intel_pt_log("No more data\n");
498                 return -ENODATA;
499         }
500         if (!buffer.consecutive) {
501                 decoder->ip = 0;
502                 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
503                 decoder->ref_timestamp = buffer.ref_timestamp;
504                 decoder->timestamp = 0;
505                 decoder->have_tma = false;
506                 decoder->state.trace_nr = buffer.trace_nr;
507                 intel_pt_log("Reference timestamp 0x%" PRIx64 "\n",
508                              decoder->ref_timestamp);
509                 return -ENOLINK;
510         }
511
512         return 0;
513 }
514
515 static int intel_pt_get_next_data(struct intel_pt_decoder *decoder)
516 {
517         if (!decoder->next_buf)
518                 return intel_pt_get_data(decoder);
519
520         decoder->buf = decoder->next_buf;
521         decoder->len = decoder->next_len;
522         decoder->next_buf = 0;
523         decoder->next_len = 0;
524         return 0;
525 }
526
527 static int intel_pt_get_split_packet(struct intel_pt_decoder *decoder)
528 {
529         unsigned char *buf = decoder->temp_buf;
530         size_t old_len, len, n;
531         int ret;
532
533         old_len = decoder->len;
534         len = decoder->len;
535         memcpy(buf, decoder->buf, len);
536
537         ret = intel_pt_get_data(decoder);
538         if (ret) {
539                 decoder->pos += old_len;
540                 return ret < 0 ? ret : -EINVAL;
541         }
542
543         n = INTEL_PT_PKT_MAX_SZ - len;
544         if (n > decoder->len)
545                 n = decoder->len;
546         memcpy(buf + len, decoder->buf, n);
547         len += n;
548
549         ret = intel_pt_get_packet(buf, len, &decoder->packet);
550         if (ret < (int)old_len) {
551                 decoder->next_buf = decoder->buf;
552                 decoder->next_len = decoder->len;
553                 decoder->buf = buf;
554                 decoder->len = old_len;
555                 return intel_pt_bad_packet(decoder);
556         }
557
558         decoder->next_buf = decoder->buf + (ret - old_len);
559         decoder->next_len = decoder->len - (ret - old_len);
560
561         decoder->buf = buf;
562         decoder->len = ret;
563
564         return ret;
565 }
566
567 struct intel_pt_pkt_info {
568         struct intel_pt_decoder   *decoder;
569         struct intel_pt_pkt       packet;
570         uint64_t                  pos;
571         int                       pkt_len;
572         int                       last_packet_type;
573         void                      *data;
574 };
575
576 typedef int (*intel_pt_pkt_cb_t)(struct intel_pt_pkt_info *pkt_info);
577
578 /* Lookahead packets in current buffer */
579 static int intel_pt_pkt_lookahead(struct intel_pt_decoder *decoder,
580                                   intel_pt_pkt_cb_t cb, void *data)
581 {
582         struct intel_pt_pkt_info pkt_info;
583         const unsigned char *buf = decoder->buf;
584         size_t len = decoder->len;
585         int ret;
586
587         pkt_info.decoder          = decoder;
588         pkt_info.pos              = decoder->pos;
589         pkt_info.pkt_len          = decoder->pkt_step;
590         pkt_info.last_packet_type = decoder->last_packet_type;
591         pkt_info.data             = data;
592
593         while (1) {
594                 do {
595                         pkt_info.pos += pkt_info.pkt_len;
596                         buf          += pkt_info.pkt_len;
597                         len          -= pkt_info.pkt_len;
598
599                         if (!len)
600                                 return INTEL_PT_NEED_MORE_BYTES;
601
602                         ret = intel_pt_get_packet(buf, len, &pkt_info.packet);
603                         if (!ret)
604                                 return INTEL_PT_NEED_MORE_BYTES;
605                         if (ret < 0)
606                                 return ret;
607
608                         pkt_info.pkt_len = ret;
609                 } while (pkt_info.packet.type == INTEL_PT_PAD);
610
611                 ret = cb(&pkt_info);
612                 if (ret)
613                         return 0;
614
615                 pkt_info.last_packet_type = pkt_info.packet.type;
616         }
617 }
618
619 struct intel_pt_calc_cyc_to_tsc_info {
620         uint64_t        cycle_cnt;
621         unsigned int    cbr;
622         uint32_t        last_mtc;
623         uint64_t        ctc_timestamp;
624         uint64_t        ctc_delta;
625         uint64_t        tsc_timestamp;
626         uint64_t        timestamp;
627         bool            have_tma;
628         bool            fixup_last_mtc;
629         bool            from_mtc;
630         double          cbr_cyc_to_tsc;
631 };
632
633 /*
634  * MTC provides a 8-bit slice of CTC but the TMA packet only provides the lower
635  * 16 bits of CTC. If mtc_shift > 8 then some of the MTC bits are not in the CTC
636  * provided by the TMA packet. Fix-up the last_mtc calculated from the TMA
637  * packet by copying the missing bits from the current MTC assuming the least
638  * difference between the two, and that the current MTC comes after last_mtc.
639  */
640 static void intel_pt_fixup_last_mtc(uint32_t mtc, int mtc_shift,
641                                     uint32_t *last_mtc)
642 {
643         uint32_t first_missing_bit = 1U << (16 - mtc_shift);
644         uint32_t mask = ~(first_missing_bit - 1);
645
646         *last_mtc |= mtc & mask;
647         if (*last_mtc >= mtc) {
648                 *last_mtc -= first_missing_bit;
649                 *last_mtc &= 0xff;
650         }
651 }
652
653 static int intel_pt_calc_cyc_cb(struct intel_pt_pkt_info *pkt_info)
654 {
655         struct intel_pt_decoder *decoder = pkt_info->decoder;
656         struct intel_pt_calc_cyc_to_tsc_info *data = pkt_info->data;
657         uint64_t timestamp;
658         double cyc_to_tsc;
659         unsigned int cbr;
660         uint32_t mtc, mtc_delta, ctc, fc, ctc_rem;
661
662         switch (pkt_info->packet.type) {
663         case INTEL_PT_TNT:
664         case INTEL_PT_TIP_PGE:
665         case INTEL_PT_TIP:
666         case INTEL_PT_FUP:
667         case INTEL_PT_PSB:
668         case INTEL_PT_PIP:
669         case INTEL_PT_MODE_EXEC:
670         case INTEL_PT_MODE_TSX:
671         case INTEL_PT_PSBEND:
672         case INTEL_PT_PAD:
673         case INTEL_PT_VMCS:
674         case INTEL_PT_MNT:
675         case INTEL_PT_PTWRITE:
676         case INTEL_PT_PTWRITE_IP:
677                 return 0;
678
679         case INTEL_PT_MTC:
680                 if (!data->have_tma)
681                         return 0;
682
683                 mtc = pkt_info->packet.payload;
684                 if (decoder->mtc_shift > 8 && data->fixup_last_mtc) {
685                         data->fixup_last_mtc = false;
686                         intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
687                                                 &data->last_mtc);
688                 }
689                 if (mtc > data->last_mtc)
690                         mtc_delta = mtc - data->last_mtc;
691                 else
692                         mtc_delta = mtc + 256 - data->last_mtc;
693                 data->ctc_delta += mtc_delta << decoder->mtc_shift;
694                 data->last_mtc = mtc;
695
696                 if (decoder->tsc_ctc_mult) {
697                         timestamp = data->ctc_timestamp +
698                                 data->ctc_delta * decoder->tsc_ctc_mult;
699                 } else {
700                         timestamp = data->ctc_timestamp +
701                                 multdiv(data->ctc_delta,
702                                         decoder->tsc_ctc_ratio_n,
703                                         decoder->tsc_ctc_ratio_d);
704                 }
705
706                 if (timestamp < data->timestamp)
707                         return 1;
708
709                 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
710                         data->timestamp = timestamp;
711                         return 0;
712                 }
713
714                 break;
715
716         case INTEL_PT_TSC:
717                 /*
718                  * For now, do not support using TSC packets - refer
719                  * intel_pt_calc_cyc_to_tsc().
720                  */
721                 if (data->from_mtc)
722                         return 1;
723                 timestamp = pkt_info->packet.payload |
724                             (data->timestamp & (0xffULL << 56));
725                 if (data->from_mtc && timestamp < data->timestamp &&
726                     data->timestamp - timestamp < decoder->tsc_slip)
727                         return 1;
728                 if (timestamp < data->timestamp)
729                         timestamp += (1ULL << 56);
730                 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
731                         if (data->from_mtc)
732                                 return 1;
733                         data->tsc_timestamp = timestamp;
734                         data->timestamp = timestamp;
735                         return 0;
736                 }
737                 break;
738
739         case INTEL_PT_TMA:
740                 if (data->from_mtc)
741                         return 1;
742
743                 if (!decoder->tsc_ctc_ratio_d)
744                         return 0;
745
746                 ctc = pkt_info->packet.payload;
747                 fc = pkt_info->packet.count;
748                 ctc_rem = ctc & decoder->ctc_rem_mask;
749
750                 data->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
751
752                 data->ctc_timestamp = data->tsc_timestamp - fc;
753                 if (decoder->tsc_ctc_mult) {
754                         data->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
755                 } else {
756                         data->ctc_timestamp -=
757                                 multdiv(ctc_rem, decoder->tsc_ctc_ratio_n,
758                                         decoder->tsc_ctc_ratio_d);
759                 }
760
761                 data->ctc_delta = 0;
762                 data->have_tma = true;
763                 data->fixup_last_mtc = true;
764
765                 return 0;
766
767         case INTEL_PT_CYC:
768                 data->cycle_cnt += pkt_info->packet.payload;
769                 return 0;
770
771         case INTEL_PT_CBR:
772                 cbr = pkt_info->packet.payload;
773                 if (data->cbr && data->cbr != cbr)
774                         return 1;
775                 data->cbr = cbr;
776                 data->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
777                 return 0;
778
779         case INTEL_PT_TIP_PGD:
780         case INTEL_PT_TRACESTOP:
781         case INTEL_PT_EXSTOP:
782         case INTEL_PT_EXSTOP_IP:
783         case INTEL_PT_MWAIT:
784         case INTEL_PT_PWRE:
785         case INTEL_PT_PWRX:
786         case INTEL_PT_OVF:
787         case INTEL_PT_BAD: /* Does not happen */
788         default:
789                 return 1;
790         }
791
792         if (!data->cbr && decoder->cbr) {
793                 data->cbr = decoder->cbr;
794                 data->cbr_cyc_to_tsc = decoder->cbr_cyc_to_tsc;
795         }
796
797         if (!data->cycle_cnt)
798                 return 1;
799
800         cyc_to_tsc = (double)(timestamp - decoder->timestamp) / data->cycle_cnt;
801
802         if (data->cbr && cyc_to_tsc > data->cbr_cyc_to_tsc &&
803             cyc_to_tsc / data->cbr_cyc_to_tsc > 1.25) {
804                 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle too big (c.f. CBR-based value %g), pos " x64_fmt "\n",
805                              cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
806                 return 1;
807         }
808
809         decoder->calc_cyc_to_tsc = cyc_to_tsc;
810         decoder->have_calc_cyc_to_tsc = true;
811
812         if (data->cbr) {
813                 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. CBR-based value %g, pos " x64_fmt "\n",
814                              cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
815         } else {
816                 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. unknown CBR-based value, pos " x64_fmt "\n",
817                              cyc_to_tsc, pkt_info->pos);
818         }
819
820         return 1;
821 }
822
823 static void intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder *decoder,
824                                      bool from_mtc)
825 {
826         struct intel_pt_calc_cyc_to_tsc_info data = {
827                 .cycle_cnt      = 0,
828                 .cbr            = 0,
829                 .last_mtc       = decoder->last_mtc,
830                 .ctc_timestamp  = decoder->ctc_timestamp,
831                 .ctc_delta      = decoder->ctc_delta,
832                 .tsc_timestamp  = decoder->tsc_timestamp,
833                 .timestamp      = decoder->timestamp,
834                 .have_tma       = decoder->have_tma,
835                 .fixup_last_mtc = decoder->fixup_last_mtc,
836                 .from_mtc       = from_mtc,
837                 .cbr_cyc_to_tsc = 0,
838         };
839
840         /*
841          * For now, do not support using TSC packets for at least the reasons:
842          * 1) timing might have stopped
843          * 2) TSC packets within PSB+ can slip against CYC packets
844          */
845         if (!from_mtc)
846                 return;
847
848         intel_pt_pkt_lookahead(decoder, intel_pt_calc_cyc_cb, &data);
849 }
850
851 static int intel_pt_get_next_packet(struct intel_pt_decoder *decoder)
852 {
853         int ret;
854
855         decoder->last_packet_type = decoder->packet.type;
856
857         do {
858                 decoder->pos += decoder->pkt_step;
859                 decoder->buf += decoder->pkt_step;
860                 decoder->len -= decoder->pkt_step;
861
862                 if (!decoder->len) {
863                         ret = intel_pt_get_next_data(decoder);
864                         if (ret)
865                                 return ret;
866                 }
867
868                 ret = intel_pt_get_packet(decoder->buf, decoder->len,
869                                           &decoder->packet);
870                 if (ret == INTEL_PT_NEED_MORE_BYTES &&
871                     decoder->len < INTEL_PT_PKT_MAX_SZ && !decoder->next_buf) {
872                         ret = intel_pt_get_split_packet(decoder);
873                         if (ret < 0)
874                                 return ret;
875                 }
876                 if (ret <= 0)
877                         return intel_pt_bad_packet(decoder);
878
879                 decoder->pkt_len = ret;
880                 decoder->pkt_step = ret;
881                 intel_pt_decoder_log_packet(decoder);
882         } while (decoder->packet.type == INTEL_PT_PAD);
883
884         return 0;
885 }
886
887 static uint64_t intel_pt_next_period(struct intel_pt_decoder *decoder)
888 {
889         uint64_t timestamp, masked_timestamp;
890
891         timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
892         masked_timestamp = timestamp & decoder->period_mask;
893         if (decoder->continuous_period) {
894                 if (masked_timestamp != decoder->last_masked_timestamp)
895                         return 1;
896         } else {
897                 timestamp += 1;
898                 masked_timestamp = timestamp & decoder->period_mask;
899                 if (masked_timestamp != decoder->last_masked_timestamp) {
900                         decoder->last_masked_timestamp = masked_timestamp;
901                         decoder->continuous_period = true;
902                 }
903         }
904         return decoder->period_ticks - (timestamp - masked_timestamp);
905 }
906
907 static uint64_t intel_pt_next_sample(struct intel_pt_decoder *decoder)
908 {
909         switch (decoder->period_type) {
910         case INTEL_PT_PERIOD_INSTRUCTIONS:
911                 return decoder->period - decoder->period_insn_cnt;
912         case INTEL_PT_PERIOD_TICKS:
913                 return intel_pt_next_period(decoder);
914         case INTEL_PT_PERIOD_NONE:
915         case INTEL_PT_PERIOD_MTC:
916         default:
917                 return 0;
918         }
919 }
920
921 static void intel_pt_sample_insn(struct intel_pt_decoder *decoder)
922 {
923         uint64_t timestamp, masked_timestamp;
924
925         switch (decoder->period_type) {
926         case INTEL_PT_PERIOD_INSTRUCTIONS:
927                 decoder->period_insn_cnt = 0;
928                 break;
929         case INTEL_PT_PERIOD_TICKS:
930                 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
931                 masked_timestamp = timestamp & decoder->period_mask;
932                 decoder->last_masked_timestamp = masked_timestamp;
933                 break;
934         case INTEL_PT_PERIOD_NONE:
935         case INTEL_PT_PERIOD_MTC:
936         default:
937                 break;
938         }
939
940         decoder->state.type |= INTEL_PT_INSTRUCTION;
941 }
942
943 static int intel_pt_walk_insn(struct intel_pt_decoder *decoder,
944                               struct intel_pt_insn *intel_pt_insn, uint64_t ip)
945 {
946         uint64_t max_insn_cnt, insn_cnt = 0;
947         int err;
948
949         if (!decoder->mtc_insn)
950                 decoder->mtc_insn = true;
951
952         max_insn_cnt = intel_pt_next_sample(decoder);
953
954         err = decoder->walk_insn(intel_pt_insn, &insn_cnt, &decoder->ip, ip,
955                                  max_insn_cnt, decoder->data);
956
957         decoder->tot_insn_cnt += insn_cnt;
958         decoder->timestamp_insn_cnt += insn_cnt;
959         decoder->sample_insn_cnt += insn_cnt;
960         decoder->period_insn_cnt += insn_cnt;
961
962         if (err) {
963                 decoder->no_progress = 0;
964                 decoder->pkt_state = INTEL_PT_STATE_ERR2;
965                 intel_pt_log_at("ERROR: Failed to get instruction",
966                                 decoder->ip);
967                 if (err == -ENOENT)
968                         return -ENOLINK;
969                 return -EILSEQ;
970         }
971
972         if (ip && decoder->ip == ip) {
973                 err = -EAGAIN;
974                 goto out;
975         }
976
977         if (max_insn_cnt && insn_cnt >= max_insn_cnt)
978                 intel_pt_sample_insn(decoder);
979
980         if (intel_pt_insn->branch == INTEL_PT_BR_NO_BRANCH) {
981                 decoder->state.type = INTEL_PT_INSTRUCTION;
982                 decoder->state.from_ip = decoder->ip;
983                 decoder->state.to_ip = 0;
984                 decoder->ip += intel_pt_insn->length;
985                 err = INTEL_PT_RETURN;
986                 goto out;
987         }
988
989         if (intel_pt_insn->op == INTEL_PT_OP_CALL) {
990                 /* Zero-length calls are excluded */
991                 if (intel_pt_insn->branch != INTEL_PT_BR_UNCONDITIONAL ||
992                     intel_pt_insn->rel) {
993                         err = intel_pt_push(&decoder->stack, decoder->ip +
994                                             intel_pt_insn->length);
995                         if (err)
996                                 goto out;
997                 }
998         } else if (intel_pt_insn->op == INTEL_PT_OP_RET) {
999                 decoder->ret_addr = intel_pt_pop(&decoder->stack);
1000         }
1001
1002         if (intel_pt_insn->branch == INTEL_PT_BR_UNCONDITIONAL) {
1003                 int cnt = decoder->no_progress++;
1004
1005                 decoder->state.from_ip = decoder->ip;
1006                 decoder->ip += intel_pt_insn->length +
1007                                 intel_pt_insn->rel;
1008                 decoder->state.to_ip = decoder->ip;
1009                 err = INTEL_PT_RETURN;
1010
1011                 /*
1012                  * Check for being stuck in a loop.  This can happen if a
1013                  * decoder error results in the decoder erroneously setting the
1014                  * ip to an address that is itself in an infinite loop that
1015                  * consumes no packets.  When that happens, there must be an
1016                  * unconditional branch.
1017                  */
1018                 if (cnt) {
1019                         if (cnt == 1) {
1020                                 decoder->stuck_ip = decoder->state.to_ip;
1021                                 decoder->stuck_ip_prd = 1;
1022                                 decoder->stuck_ip_cnt = 1;
1023                         } else if (cnt > INTEL_PT_MAX_LOOPS ||
1024                                    decoder->state.to_ip == decoder->stuck_ip) {
1025                                 intel_pt_log_at("ERROR: Never-ending loop",
1026                                                 decoder->state.to_ip);
1027                                 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1028                                 err = -ELOOP;
1029                                 goto out;
1030                         } else if (!--decoder->stuck_ip_cnt) {
1031                                 decoder->stuck_ip_prd += 1;
1032                                 decoder->stuck_ip_cnt = decoder->stuck_ip_prd;
1033                                 decoder->stuck_ip = decoder->state.to_ip;
1034                         }
1035                 }
1036                 goto out_no_progress;
1037         }
1038 out:
1039         decoder->no_progress = 0;
1040 out_no_progress:
1041         decoder->state.insn_op = intel_pt_insn->op;
1042         decoder->state.insn_len = intel_pt_insn->length;
1043         memcpy(decoder->state.insn, intel_pt_insn->buf,
1044                INTEL_PT_INSN_BUF_SZ);
1045
1046         if (decoder->tx_flags & INTEL_PT_IN_TX)
1047                 decoder->state.flags |= INTEL_PT_IN_TX;
1048
1049         return err;
1050 }
1051
1052 static bool intel_pt_fup_event(struct intel_pt_decoder *decoder)
1053 {
1054         bool ret = false;
1055
1056         if (decoder->set_fup_tx_flags) {
1057                 decoder->set_fup_tx_flags = false;
1058                 decoder->tx_flags = decoder->fup_tx_flags;
1059                 decoder->state.type = INTEL_PT_TRANSACTION;
1060                 decoder->state.from_ip = decoder->ip;
1061                 decoder->state.to_ip = 0;
1062                 decoder->state.flags = decoder->fup_tx_flags;
1063                 return true;
1064         }
1065         if (decoder->set_fup_ptw) {
1066                 decoder->set_fup_ptw = false;
1067                 decoder->state.type = INTEL_PT_PTW;
1068                 decoder->state.flags |= INTEL_PT_FUP_IP;
1069                 decoder->state.from_ip = decoder->ip;
1070                 decoder->state.to_ip = 0;
1071                 decoder->state.ptw_payload = decoder->fup_ptw_payload;
1072                 return true;
1073         }
1074         if (decoder->set_fup_mwait) {
1075                 decoder->set_fup_mwait = false;
1076                 decoder->state.type = INTEL_PT_MWAIT_OP;
1077                 decoder->state.from_ip = decoder->ip;
1078                 decoder->state.to_ip = 0;
1079                 decoder->state.mwait_payload = decoder->fup_mwait_payload;
1080                 ret = true;
1081         }
1082         if (decoder->set_fup_pwre) {
1083                 decoder->set_fup_pwre = false;
1084                 decoder->state.type |= INTEL_PT_PWR_ENTRY;
1085                 decoder->state.type &= ~INTEL_PT_BRANCH;
1086                 decoder->state.from_ip = decoder->ip;
1087                 decoder->state.to_ip = 0;
1088                 decoder->state.pwre_payload = decoder->fup_pwre_payload;
1089                 ret = true;
1090         }
1091         if (decoder->set_fup_exstop) {
1092                 decoder->set_fup_exstop = false;
1093                 decoder->state.type |= INTEL_PT_EX_STOP;
1094                 decoder->state.type &= ~INTEL_PT_BRANCH;
1095                 decoder->state.flags |= INTEL_PT_FUP_IP;
1096                 decoder->state.from_ip = decoder->ip;
1097                 decoder->state.to_ip = 0;
1098                 ret = true;
1099         }
1100         return ret;
1101 }
1102
1103 static inline bool intel_pt_fup_with_nlip(struct intel_pt_decoder *decoder,
1104                                           struct intel_pt_insn *intel_pt_insn,
1105                                           uint64_t ip, int err)
1106 {
1107         return decoder->flags & INTEL_PT_FUP_WITH_NLIP && !err &&
1108                intel_pt_insn->branch == INTEL_PT_BR_INDIRECT &&
1109                ip == decoder->ip + intel_pt_insn->length;
1110 }
1111
1112 static int intel_pt_walk_fup(struct intel_pt_decoder *decoder)
1113 {
1114         struct intel_pt_insn intel_pt_insn;
1115         uint64_t ip;
1116         int err;
1117
1118         ip = decoder->last_ip;
1119
1120         while (1) {
1121                 err = intel_pt_walk_insn(decoder, &intel_pt_insn, ip);
1122                 if (err == INTEL_PT_RETURN)
1123                         return 0;
1124                 if (err == -EAGAIN ||
1125                     intel_pt_fup_with_nlip(decoder, &intel_pt_insn, ip, err)) {
1126                         if (intel_pt_fup_event(decoder))
1127                                 return 0;
1128                         return -EAGAIN;
1129                 }
1130                 decoder->set_fup_tx_flags = false;
1131                 if (err)
1132                         return err;
1133
1134                 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1135                         intel_pt_log_at("ERROR: Unexpected indirect branch",
1136                                         decoder->ip);
1137                         decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1138                         return -ENOENT;
1139                 }
1140
1141                 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1142                         intel_pt_log_at("ERROR: Unexpected conditional branch",
1143                                         decoder->ip);
1144                         decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1145                         return -ENOENT;
1146                 }
1147
1148                 intel_pt_bug(decoder);
1149         }
1150 }
1151
1152 static int intel_pt_walk_tip(struct intel_pt_decoder *decoder)
1153 {
1154         struct intel_pt_insn intel_pt_insn;
1155         int err;
1156
1157         err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1158         if (err == INTEL_PT_RETURN &&
1159             decoder->pgd_ip &&
1160             decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1161             (decoder->state.type & INTEL_PT_BRANCH) &&
1162             decoder->pgd_ip(decoder->state.to_ip, decoder->data)) {
1163                 /* Unconditional branch leaving filter region */
1164                 decoder->no_progress = 0;
1165                 decoder->pge = false;
1166                 decoder->continuous_period = false;
1167                 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1168                 decoder->state.type |= INTEL_PT_TRACE_END;
1169                 return 0;
1170         }
1171         if (err == INTEL_PT_RETURN)
1172                 return 0;
1173         if (err)
1174                 return err;
1175
1176         if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1177                 if (decoder->pkt_state == INTEL_PT_STATE_TIP_PGD) {
1178                         decoder->pge = false;
1179                         decoder->continuous_period = false;
1180                         decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1181                         decoder->state.from_ip = decoder->ip;
1182                         if (decoder->packet.count == 0) {
1183                                 decoder->state.to_ip = 0;
1184                         } else {
1185                                 decoder->state.to_ip = decoder->last_ip;
1186                                 decoder->ip = decoder->last_ip;
1187                         }
1188                         decoder->state.type |= INTEL_PT_TRACE_END;
1189                 } else {
1190                         decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1191                         decoder->state.from_ip = decoder->ip;
1192                         if (decoder->packet.count == 0) {
1193                                 decoder->state.to_ip = 0;
1194                         } else {
1195                                 decoder->state.to_ip = decoder->last_ip;
1196                                 decoder->ip = decoder->last_ip;
1197                         }
1198                 }
1199                 return 0;
1200         }
1201
1202         if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1203                 uint64_t to_ip = decoder->ip + intel_pt_insn.length +
1204                                  intel_pt_insn.rel;
1205
1206                 if (decoder->pgd_ip &&
1207                     decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1208                     decoder->pgd_ip(to_ip, decoder->data)) {
1209                         /* Conditional branch leaving filter region */
1210                         decoder->pge = false;
1211                         decoder->continuous_period = false;
1212                         decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1213                         decoder->ip = to_ip;
1214                         decoder->state.from_ip = decoder->ip;
1215                         decoder->state.to_ip = to_ip;
1216                         decoder->state.type |= INTEL_PT_TRACE_END;
1217                         return 0;
1218                 }
1219                 intel_pt_log_at("ERROR: Conditional branch when expecting indirect branch",
1220                                 decoder->ip);
1221                 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1222                 return -ENOENT;
1223         }
1224
1225         return intel_pt_bug(decoder);
1226 }
1227
1228 static int intel_pt_walk_tnt(struct intel_pt_decoder *decoder)
1229 {
1230         struct intel_pt_insn intel_pt_insn;
1231         int err;
1232
1233         while (1) {
1234                 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1235                 if (err == INTEL_PT_RETURN)
1236                         return 0;
1237                 if (err)
1238                         return err;
1239
1240                 if (intel_pt_insn.op == INTEL_PT_OP_RET) {
1241                         if (!decoder->return_compression) {
1242                                 intel_pt_log_at("ERROR: RET when expecting conditional branch",
1243                                                 decoder->ip);
1244                                 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1245                                 return -ENOENT;
1246                         }
1247                         if (!decoder->ret_addr) {
1248                                 intel_pt_log_at("ERROR: Bad RET compression (stack empty)",
1249                                                 decoder->ip);
1250                                 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1251                                 return -ENOENT;
1252                         }
1253                         if (!(decoder->tnt.payload & BIT63)) {
1254                                 intel_pt_log_at("ERROR: Bad RET compression (TNT=N)",
1255                                                 decoder->ip);
1256                                 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1257                                 return -ENOENT;
1258                         }
1259                         decoder->tnt.count -= 1;
1260                         if (!decoder->tnt.count)
1261                                 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1262                         decoder->tnt.payload <<= 1;
1263                         decoder->state.from_ip = decoder->ip;
1264                         decoder->ip = decoder->ret_addr;
1265                         decoder->state.to_ip = decoder->ip;
1266                         return 0;
1267                 }
1268
1269                 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1270                         /* Handle deferred TIPs */
1271                         err = intel_pt_get_next_packet(decoder);
1272                         if (err)
1273                                 return err;
1274                         if (decoder->packet.type != INTEL_PT_TIP ||
1275                             decoder->packet.count == 0) {
1276                                 intel_pt_log_at("ERROR: Missing deferred TIP for indirect branch",
1277                                                 decoder->ip);
1278                                 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1279                                 decoder->pkt_step = 0;
1280                                 return -ENOENT;
1281                         }
1282                         intel_pt_set_last_ip(decoder);
1283                         decoder->state.from_ip = decoder->ip;
1284                         decoder->state.to_ip = decoder->last_ip;
1285                         decoder->ip = decoder->last_ip;
1286                         return 0;
1287                 }
1288
1289                 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1290                         decoder->tnt.count -= 1;
1291                         if (!decoder->tnt.count)
1292                                 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1293                         if (decoder->tnt.payload & BIT63) {
1294                                 decoder->tnt.payload <<= 1;
1295                                 decoder->state.from_ip = decoder->ip;
1296                                 decoder->ip += intel_pt_insn.length +
1297                                                intel_pt_insn.rel;
1298                                 decoder->state.to_ip = decoder->ip;
1299                                 return 0;
1300                         }
1301                         /* Instruction sample for a non-taken branch */
1302                         if (decoder->state.type & INTEL_PT_INSTRUCTION) {
1303                                 decoder->tnt.payload <<= 1;
1304                                 decoder->state.type = INTEL_PT_INSTRUCTION;
1305                                 decoder->state.from_ip = decoder->ip;
1306                                 decoder->state.to_ip = 0;
1307                                 decoder->ip += intel_pt_insn.length;
1308                                 return 0;
1309                         }
1310                         decoder->ip += intel_pt_insn.length;
1311                         if (!decoder->tnt.count)
1312                                 return -EAGAIN;
1313                         decoder->tnt.payload <<= 1;
1314                         continue;
1315                 }
1316
1317                 return intel_pt_bug(decoder);
1318         }
1319 }
1320
1321 static int intel_pt_mode_tsx(struct intel_pt_decoder *decoder, bool *no_tip)
1322 {
1323         unsigned int fup_tx_flags;
1324         int err;
1325
1326         fup_tx_flags = decoder->packet.payload &
1327                        (INTEL_PT_IN_TX | INTEL_PT_ABORT_TX);
1328         err = intel_pt_get_next_packet(decoder);
1329         if (err)
1330                 return err;
1331         if (decoder->packet.type == INTEL_PT_FUP) {
1332                 decoder->fup_tx_flags = fup_tx_flags;
1333                 decoder->set_fup_tx_flags = true;
1334                 if (!(decoder->fup_tx_flags & INTEL_PT_ABORT_TX))
1335                         *no_tip = true;
1336         } else {
1337                 intel_pt_log_at("ERROR: Missing FUP after MODE.TSX",
1338                                 decoder->pos);
1339                 intel_pt_update_in_tx(decoder);
1340         }
1341         return 0;
1342 }
1343
1344 static void intel_pt_calc_tsc_timestamp(struct intel_pt_decoder *decoder)
1345 {
1346         uint64_t timestamp;
1347
1348         decoder->have_tma = false;
1349
1350         if (decoder->ref_timestamp) {
1351                 timestamp = decoder->packet.payload |
1352                             (decoder->ref_timestamp & (0xffULL << 56));
1353                 if (timestamp < decoder->ref_timestamp) {
1354                         if (decoder->ref_timestamp - timestamp > (1ULL << 55))
1355                                 timestamp += (1ULL << 56);
1356                 } else {
1357                         if (timestamp - decoder->ref_timestamp > (1ULL << 55))
1358                                 timestamp -= (1ULL << 56);
1359                 }
1360                 decoder->tsc_timestamp = timestamp;
1361                 decoder->timestamp = timestamp;
1362                 decoder->ref_timestamp = 0;
1363                 decoder->timestamp_insn_cnt = 0;
1364         } else if (decoder->timestamp) {
1365                 timestamp = decoder->packet.payload |
1366                             (decoder->timestamp & (0xffULL << 56));
1367                 decoder->tsc_timestamp = timestamp;
1368                 if (timestamp < decoder->timestamp &&
1369                     decoder->timestamp - timestamp < decoder->tsc_slip) {
1370                         intel_pt_log_to("Suppressing backwards timestamp",
1371                                         timestamp);
1372                         timestamp = decoder->timestamp;
1373                 }
1374                 if (timestamp < decoder->timestamp) {
1375                         intel_pt_log_to("Wraparound timestamp", timestamp);
1376                         timestamp += (1ULL << 56);
1377                         decoder->tsc_timestamp = timestamp;
1378                 }
1379                 decoder->timestamp = timestamp;
1380                 decoder->timestamp_insn_cnt = 0;
1381         }
1382
1383         if (decoder->last_packet_type == INTEL_PT_CYC) {
1384                 decoder->cyc_ref_timestamp = decoder->timestamp;
1385                 decoder->cycle_cnt = 0;
1386                 decoder->have_calc_cyc_to_tsc = false;
1387                 intel_pt_calc_cyc_to_tsc(decoder, false);
1388         }
1389
1390         intel_pt_log_to("Setting timestamp", decoder->timestamp);
1391 }
1392
1393 static int intel_pt_overflow(struct intel_pt_decoder *decoder)
1394 {
1395         intel_pt_log("ERROR: Buffer overflow\n");
1396         intel_pt_clear_tx_flags(decoder);
1397         decoder->cbr = 0;
1398         decoder->timestamp_insn_cnt = 0;
1399         decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1400         decoder->overflow = true;
1401         return -EOVERFLOW;
1402 }
1403
1404 static void intel_pt_calc_tma(struct intel_pt_decoder *decoder)
1405 {
1406         uint32_t ctc = decoder->packet.payload;
1407         uint32_t fc = decoder->packet.count;
1408         uint32_t ctc_rem = ctc & decoder->ctc_rem_mask;
1409
1410         if (!decoder->tsc_ctc_ratio_d)
1411                 return;
1412
1413         decoder->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
1414         decoder->ctc_timestamp = decoder->tsc_timestamp - fc;
1415         if (decoder->tsc_ctc_mult) {
1416                 decoder->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
1417         } else {
1418                 decoder->ctc_timestamp -= multdiv(ctc_rem,
1419                                                   decoder->tsc_ctc_ratio_n,
1420                                                   decoder->tsc_ctc_ratio_d);
1421         }
1422         decoder->ctc_delta = 0;
1423         decoder->have_tma = true;
1424         decoder->fixup_last_mtc = true;
1425         intel_pt_log("CTC timestamp " x64_fmt " last MTC %#x  CTC rem %#x\n",
1426                      decoder->ctc_timestamp, decoder->last_mtc, ctc_rem);
1427 }
1428
1429 static void intel_pt_calc_mtc_timestamp(struct intel_pt_decoder *decoder)
1430 {
1431         uint64_t timestamp;
1432         uint32_t mtc, mtc_delta;
1433
1434         if (!decoder->have_tma)
1435                 return;
1436
1437         mtc = decoder->packet.payload;
1438
1439         if (decoder->mtc_shift > 8 && decoder->fixup_last_mtc) {
1440                 decoder->fixup_last_mtc = false;
1441                 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
1442                                         &decoder->last_mtc);
1443         }
1444
1445         if (mtc > decoder->last_mtc)
1446                 mtc_delta = mtc - decoder->last_mtc;
1447         else
1448                 mtc_delta = mtc + 256 - decoder->last_mtc;
1449
1450         decoder->ctc_delta += mtc_delta << decoder->mtc_shift;
1451
1452         if (decoder->tsc_ctc_mult) {
1453                 timestamp = decoder->ctc_timestamp +
1454                             decoder->ctc_delta * decoder->tsc_ctc_mult;
1455         } else {
1456                 timestamp = decoder->ctc_timestamp +
1457                             multdiv(decoder->ctc_delta,
1458                                     decoder->tsc_ctc_ratio_n,
1459                                     decoder->tsc_ctc_ratio_d);
1460         }
1461
1462         if (timestamp < decoder->timestamp)
1463                 intel_pt_log("Suppressing MTC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1464                              timestamp, decoder->timestamp);
1465         else
1466                 decoder->timestamp = timestamp;
1467
1468         decoder->timestamp_insn_cnt = 0;
1469         decoder->last_mtc = mtc;
1470
1471         if (decoder->last_packet_type == INTEL_PT_CYC) {
1472                 decoder->cyc_ref_timestamp = decoder->timestamp;
1473                 decoder->cycle_cnt = 0;
1474                 decoder->have_calc_cyc_to_tsc = false;
1475                 intel_pt_calc_cyc_to_tsc(decoder, true);
1476         }
1477 }
1478
1479 static void intel_pt_calc_cbr(struct intel_pt_decoder *decoder)
1480 {
1481         unsigned int cbr = decoder->packet.payload & 0xff;
1482
1483         decoder->cbr_payload = decoder->packet.payload;
1484
1485         if (decoder->cbr == cbr)
1486                 return;
1487
1488         decoder->cbr = cbr;
1489         decoder->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
1490 }
1491
1492 static void intel_pt_calc_cyc_timestamp(struct intel_pt_decoder *decoder)
1493 {
1494         uint64_t timestamp = decoder->cyc_ref_timestamp;
1495
1496         decoder->have_cyc = true;
1497
1498         decoder->cycle_cnt += decoder->packet.payload;
1499
1500         if (!decoder->cyc_ref_timestamp)
1501                 return;
1502
1503         if (decoder->have_calc_cyc_to_tsc)
1504                 timestamp += decoder->cycle_cnt * decoder->calc_cyc_to_tsc;
1505         else if (decoder->cbr)
1506                 timestamp += decoder->cycle_cnt * decoder->cbr_cyc_to_tsc;
1507         else
1508                 return;
1509
1510         if (timestamp < decoder->timestamp)
1511                 intel_pt_log("Suppressing CYC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1512                              timestamp, decoder->timestamp);
1513         else
1514                 decoder->timestamp = timestamp;
1515
1516         decoder->timestamp_insn_cnt = 0;
1517 }
1518
1519 /* Walk PSB+ packets when already in sync. */
1520 static int intel_pt_walk_psbend(struct intel_pt_decoder *decoder)
1521 {
1522         int err;
1523
1524         while (1) {
1525                 err = intel_pt_get_next_packet(decoder);
1526                 if (err)
1527                         return err;
1528
1529                 switch (decoder->packet.type) {
1530                 case INTEL_PT_PSBEND:
1531                         return 0;
1532
1533                 case INTEL_PT_TIP_PGD:
1534                 case INTEL_PT_TIP_PGE:
1535                 case INTEL_PT_TIP:
1536                 case INTEL_PT_TNT:
1537                 case INTEL_PT_TRACESTOP:
1538                 case INTEL_PT_BAD:
1539                 case INTEL_PT_PSB:
1540                 case INTEL_PT_PTWRITE:
1541                 case INTEL_PT_PTWRITE_IP:
1542                 case INTEL_PT_EXSTOP:
1543                 case INTEL_PT_EXSTOP_IP:
1544                 case INTEL_PT_MWAIT:
1545                 case INTEL_PT_PWRE:
1546                 case INTEL_PT_PWRX:
1547                         decoder->have_tma = false;
1548                         intel_pt_log("ERROR: Unexpected packet\n");
1549                         return -EAGAIN;
1550
1551                 case INTEL_PT_OVF:
1552                         return intel_pt_overflow(decoder);
1553
1554                 case INTEL_PT_TSC:
1555                         intel_pt_calc_tsc_timestamp(decoder);
1556                         break;
1557
1558                 case INTEL_PT_TMA:
1559                         intel_pt_calc_tma(decoder);
1560                         break;
1561
1562                 case INTEL_PT_CBR:
1563                         intel_pt_calc_cbr(decoder);
1564                         break;
1565
1566                 case INTEL_PT_MODE_EXEC:
1567                         decoder->exec_mode = decoder->packet.payload;
1568                         break;
1569
1570                 case INTEL_PT_PIP:
1571                         decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1572                         break;
1573
1574                 case INTEL_PT_FUP:
1575                         decoder->pge = true;
1576                         if (decoder->packet.count)
1577                                 intel_pt_set_last_ip(decoder);
1578                         break;
1579
1580                 case INTEL_PT_MODE_TSX:
1581                         intel_pt_update_in_tx(decoder);
1582                         break;
1583
1584                 case INTEL_PT_MTC:
1585                         intel_pt_calc_mtc_timestamp(decoder);
1586                         if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1587                                 decoder->state.type |= INTEL_PT_INSTRUCTION;
1588                         break;
1589
1590                 case INTEL_PT_CYC:
1591                 case INTEL_PT_VMCS:
1592                 case INTEL_PT_MNT:
1593                 case INTEL_PT_PAD:
1594                 default:
1595                         break;
1596                 }
1597         }
1598 }
1599
1600 static int intel_pt_walk_fup_tip(struct intel_pt_decoder *decoder)
1601 {
1602         int err;
1603
1604         if (decoder->tx_flags & INTEL_PT_ABORT_TX) {
1605                 decoder->tx_flags = 0;
1606                 decoder->state.flags &= ~INTEL_PT_IN_TX;
1607                 decoder->state.flags |= INTEL_PT_ABORT_TX;
1608         } else {
1609                 decoder->state.flags |= INTEL_PT_ASYNC;
1610         }
1611
1612         while (1) {
1613                 err = intel_pt_get_next_packet(decoder);
1614                 if (err)
1615                         return err;
1616
1617                 switch (decoder->packet.type) {
1618                 case INTEL_PT_TNT:
1619                 case INTEL_PT_FUP:
1620                 case INTEL_PT_TRACESTOP:
1621                 case INTEL_PT_PSB:
1622                 case INTEL_PT_TSC:
1623                 case INTEL_PT_TMA:
1624                 case INTEL_PT_MODE_TSX:
1625                 case INTEL_PT_BAD:
1626                 case INTEL_PT_PSBEND:
1627                 case INTEL_PT_PTWRITE:
1628                 case INTEL_PT_PTWRITE_IP:
1629                 case INTEL_PT_EXSTOP:
1630                 case INTEL_PT_EXSTOP_IP:
1631                 case INTEL_PT_MWAIT:
1632                 case INTEL_PT_PWRE:
1633                 case INTEL_PT_PWRX:
1634                         intel_pt_log("ERROR: Missing TIP after FUP\n");
1635                         decoder->pkt_state = INTEL_PT_STATE_ERR3;
1636                         decoder->pkt_step = 0;
1637                         return -ENOENT;
1638
1639                 case INTEL_PT_CBR:
1640                         intel_pt_calc_cbr(decoder);
1641                         break;
1642
1643                 case INTEL_PT_OVF:
1644                         return intel_pt_overflow(decoder);
1645
1646                 case INTEL_PT_TIP_PGD:
1647                         decoder->state.from_ip = decoder->ip;
1648                         if (decoder->packet.count == 0) {
1649                                 decoder->state.to_ip = 0;
1650                         } else {
1651                                 intel_pt_set_ip(decoder);
1652                                 decoder->state.to_ip = decoder->ip;
1653                         }
1654                         decoder->pge = false;
1655                         decoder->continuous_period = false;
1656                         decoder->state.type |= INTEL_PT_TRACE_END;
1657                         return 0;
1658
1659                 case INTEL_PT_TIP_PGE:
1660                         decoder->pge = true;
1661                         intel_pt_log("Omitting PGE ip " x64_fmt "\n",
1662                                      decoder->ip);
1663                         decoder->state.from_ip = 0;
1664                         if (decoder->packet.count == 0) {
1665                                 decoder->state.to_ip = 0;
1666                         } else {
1667                                 intel_pt_set_ip(decoder);
1668                                 decoder->state.to_ip = decoder->ip;
1669                         }
1670                         decoder->state.type |= INTEL_PT_TRACE_BEGIN;
1671                         return 0;
1672
1673                 case INTEL_PT_TIP:
1674                         decoder->state.from_ip = decoder->ip;
1675                         if (decoder->packet.count == 0) {
1676                                 decoder->state.to_ip = 0;
1677                         } else {
1678                                 intel_pt_set_ip(decoder);
1679                                 decoder->state.to_ip = decoder->ip;
1680                         }
1681                         return 0;
1682
1683                 case INTEL_PT_PIP:
1684                         decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1685                         break;
1686
1687                 case INTEL_PT_MTC:
1688                         intel_pt_calc_mtc_timestamp(decoder);
1689                         if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1690                                 decoder->state.type |= INTEL_PT_INSTRUCTION;
1691                         break;
1692
1693                 case INTEL_PT_CYC:
1694                         intel_pt_calc_cyc_timestamp(decoder);
1695                         break;
1696
1697                 case INTEL_PT_MODE_EXEC:
1698                         decoder->exec_mode = decoder->packet.payload;
1699                         break;
1700
1701                 case INTEL_PT_VMCS:
1702                 case INTEL_PT_MNT:
1703                 case INTEL_PT_PAD:
1704                         break;
1705
1706                 default:
1707                         return intel_pt_bug(decoder);
1708                 }
1709         }
1710 }
1711
1712 static int intel_pt_walk_trace(struct intel_pt_decoder *decoder)
1713 {
1714         bool no_tip = false;
1715         int err;
1716
1717         while (1) {
1718                 err = intel_pt_get_next_packet(decoder);
1719                 if (err)
1720                         return err;
1721 next:
1722                 switch (decoder->packet.type) {
1723                 case INTEL_PT_TNT:
1724                         if (!decoder->packet.count)
1725                                 break;
1726                         decoder->tnt = decoder->packet;
1727                         decoder->pkt_state = INTEL_PT_STATE_TNT;
1728                         err = intel_pt_walk_tnt(decoder);
1729                         if (err == -EAGAIN)
1730                                 break;
1731                         return err;
1732
1733                 case INTEL_PT_TIP_PGD:
1734                         if (decoder->packet.count != 0)
1735                                 intel_pt_set_last_ip(decoder);
1736                         decoder->pkt_state = INTEL_PT_STATE_TIP_PGD;
1737                         return intel_pt_walk_tip(decoder);
1738
1739                 case INTEL_PT_TIP_PGE: {
1740                         decoder->pge = true;
1741                         if (decoder->packet.count == 0) {
1742                                 intel_pt_log_at("Skipping zero TIP.PGE",
1743                                                 decoder->pos);
1744                                 break;
1745                         }
1746                         intel_pt_set_ip(decoder);
1747                         decoder->state.from_ip = 0;
1748                         decoder->state.to_ip = decoder->ip;
1749                         decoder->state.type |= INTEL_PT_TRACE_BEGIN;
1750                         return 0;
1751                 }
1752
1753                 case INTEL_PT_OVF:
1754                         return intel_pt_overflow(decoder);
1755
1756                 case INTEL_PT_TIP:
1757                         if (decoder->packet.count != 0)
1758                                 intel_pt_set_last_ip(decoder);
1759                         decoder->pkt_state = INTEL_PT_STATE_TIP;
1760                         return intel_pt_walk_tip(decoder);
1761
1762                 case INTEL_PT_FUP:
1763                         if (decoder->packet.count == 0) {
1764                                 intel_pt_log_at("Skipping zero FUP",
1765                                                 decoder->pos);
1766                                 no_tip = false;
1767                                 break;
1768                         }
1769                         intel_pt_set_last_ip(decoder);
1770                         if (!decoder->branch_enable) {
1771                                 decoder->ip = decoder->last_ip;
1772                                 if (intel_pt_fup_event(decoder))
1773                                         return 0;
1774                                 no_tip = false;
1775                                 break;
1776                         }
1777                         if (decoder->set_fup_mwait)
1778                                 no_tip = true;
1779                         err = intel_pt_walk_fup(decoder);
1780                         if (err != -EAGAIN) {
1781                                 if (err)
1782                                         return err;
1783                                 if (no_tip)
1784                                         decoder->pkt_state =
1785                                                 INTEL_PT_STATE_FUP_NO_TIP;
1786                                 else
1787                                         decoder->pkt_state = INTEL_PT_STATE_FUP;
1788                                 return 0;
1789                         }
1790                         if (no_tip) {
1791                                 no_tip = false;
1792                                 break;
1793                         }
1794                         return intel_pt_walk_fup_tip(decoder);
1795
1796                 case INTEL_PT_TRACESTOP:
1797                         decoder->pge = false;
1798                         decoder->continuous_period = false;
1799                         intel_pt_clear_tx_flags(decoder);
1800                         decoder->have_tma = false;
1801                         break;
1802
1803                 case INTEL_PT_PSB:
1804                         decoder->last_ip = 0;
1805                         decoder->have_last_ip = true;
1806                         intel_pt_clear_stack(&decoder->stack);
1807                         err = intel_pt_walk_psbend(decoder);
1808                         if (err == -EAGAIN)
1809                                 goto next;
1810                         if (err)
1811                                 return err;
1812                         break;
1813
1814                 case INTEL_PT_PIP:
1815                         decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1816                         break;
1817
1818                 case INTEL_PT_MTC:
1819                         intel_pt_calc_mtc_timestamp(decoder);
1820                         if (decoder->period_type != INTEL_PT_PERIOD_MTC)
1821                                 break;
1822                         /*
1823                          * Ensure that there has been an instruction since the
1824                          * last MTC.
1825                          */
1826                         if (!decoder->mtc_insn)
1827                                 break;
1828                         decoder->mtc_insn = false;
1829                         /* Ensure that there is a timestamp */
1830                         if (!decoder->timestamp)
1831                                 break;
1832                         decoder->state.type = INTEL_PT_INSTRUCTION;
1833                         decoder->state.from_ip = decoder->ip;
1834                         decoder->state.to_ip = 0;
1835                         decoder->mtc_insn = false;
1836                         return 0;
1837
1838                 case INTEL_PT_TSC:
1839                         intel_pt_calc_tsc_timestamp(decoder);
1840                         break;
1841
1842                 case INTEL_PT_TMA:
1843                         intel_pt_calc_tma(decoder);
1844                         break;
1845
1846                 case INTEL_PT_CYC:
1847                         intel_pt_calc_cyc_timestamp(decoder);
1848                         break;
1849
1850                 case INTEL_PT_CBR:
1851                         intel_pt_calc_cbr(decoder);
1852                         if (!decoder->branch_enable &&
1853                             decoder->cbr != decoder->cbr_seen) {
1854                                 decoder->cbr_seen = decoder->cbr;
1855                                 decoder->state.type = INTEL_PT_CBR_CHG;
1856                                 decoder->state.from_ip = decoder->ip;
1857                                 decoder->state.to_ip = 0;
1858                                 decoder->state.cbr_payload =
1859                                                         decoder->packet.payload;
1860                                 return 0;
1861                         }
1862                         break;
1863
1864                 case INTEL_PT_MODE_EXEC:
1865                         decoder->exec_mode = decoder->packet.payload;
1866                         break;
1867
1868                 case INTEL_PT_MODE_TSX:
1869                         /* MODE_TSX need not be followed by FUP */
1870                         if (!decoder->pge) {
1871                                 intel_pt_update_in_tx(decoder);
1872                                 break;
1873                         }
1874                         err = intel_pt_mode_tsx(decoder, &no_tip);
1875                         if (err)
1876                                 return err;
1877                         goto next;
1878
1879                 case INTEL_PT_BAD: /* Does not happen */
1880                         return intel_pt_bug(decoder);
1881
1882                 case INTEL_PT_PSBEND:
1883                 case INTEL_PT_VMCS:
1884                 case INTEL_PT_MNT:
1885                 case INTEL_PT_PAD:
1886                         break;
1887
1888                 case INTEL_PT_PTWRITE_IP:
1889                         decoder->fup_ptw_payload = decoder->packet.payload;
1890                         err = intel_pt_get_next_packet(decoder);
1891                         if (err)
1892                                 return err;
1893                         if (decoder->packet.type == INTEL_PT_FUP) {
1894                                 decoder->set_fup_ptw = true;
1895                                 no_tip = true;
1896                         } else {
1897                                 intel_pt_log_at("ERROR: Missing FUP after PTWRITE",
1898                                                 decoder->pos);
1899                         }
1900                         goto next;
1901
1902                 case INTEL_PT_PTWRITE:
1903                         decoder->state.type = INTEL_PT_PTW;
1904                         decoder->state.from_ip = decoder->ip;
1905                         decoder->state.to_ip = 0;
1906                         decoder->state.ptw_payload = decoder->packet.payload;
1907                         return 0;
1908
1909                 case INTEL_PT_MWAIT:
1910                         decoder->fup_mwait_payload = decoder->packet.payload;
1911                         decoder->set_fup_mwait = true;
1912                         break;
1913
1914                 case INTEL_PT_PWRE:
1915                         if (decoder->set_fup_mwait) {
1916                                 decoder->fup_pwre_payload =
1917                                                         decoder->packet.payload;
1918                                 decoder->set_fup_pwre = true;
1919                                 break;
1920                         }
1921                         decoder->state.type = INTEL_PT_PWR_ENTRY;
1922                         decoder->state.from_ip = decoder->ip;
1923                         decoder->state.to_ip = 0;
1924                         decoder->state.pwrx_payload = decoder->packet.payload;
1925                         return 0;
1926
1927                 case INTEL_PT_EXSTOP_IP:
1928                         err = intel_pt_get_next_packet(decoder);
1929                         if (err)
1930                                 return err;
1931                         if (decoder->packet.type == INTEL_PT_FUP) {
1932                                 decoder->set_fup_exstop = true;
1933                                 no_tip = true;
1934                         } else {
1935                                 intel_pt_log_at("ERROR: Missing FUP after EXSTOP",
1936                                                 decoder->pos);
1937                         }
1938                         goto next;
1939
1940                 case INTEL_PT_EXSTOP:
1941                         decoder->state.type = INTEL_PT_EX_STOP;
1942                         decoder->state.from_ip = decoder->ip;
1943                         decoder->state.to_ip = 0;
1944                         return 0;
1945
1946                 case INTEL_PT_PWRX:
1947                         decoder->state.type = INTEL_PT_PWR_EXIT;
1948                         decoder->state.from_ip = decoder->ip;
1949                         decoder->state.to_ip = 0;
1950                         decoder->state.pwrx_payload = decoder->packet.payload;
1951                         return 0;
1952
1953                 default:
1954                         return intel_pt_bug(decoder);
1955                 }
1956         }
1957 }
1958
1959 static inline bool intel_pt_have_ip(struct intel_pt_decoder *decoder)
1960 {
1961         return decoder->packet.count &&
1962                (decoder->have_last_ip || decoder->packet.count == 3 ||
1963                 decoder->packet.count == 6);
1964 }
1965
1966 /* Walk PSB+ packets to get in sync. */
1967 static int intel_pt_walk_psb(struct intel_pt_decoder *decoder)
1968 {
1969         int err;
1970
1971         while (1) {
1972                 err = intel_pt_get_next_packet(decoder);
1973                 if (err)
1974                         return err;
1975
1976                 switch (decoder->packet.type) {
1977                 case INTEL_PT_TIP_PGD:
1978                         decoder->continuous_period = false;
1979                         __fallthrough;
1980                 case INTEL_PT_TIP_PGE:
1981                 case INTEL_PT_TIP:
1982                 case INTEL_PT_PTWRITE:
1983                 case INTEL_PT_PTWRITE_IP:
1984                 case INTEL_PT_EXSTOP:
1985                 case INTEL_PT_EXSTOP_IP:
1986                 case INTEL_PT_MWAIT:
1987                 case INTEL_PT_PWRE:
1988                 case INTEL_PT_PWRX:
1989                         intel_pt_log("ERROR: Unexpected packet\n");
1990                         return -ENOENT;
1991
1992                 case INTEL_PT_FUP:
1993                         decoder->pge = true;
1994                         if (intel_pt_have_ip(decoder)) {
1995                                 uint64_t current_ip = decoder->ip;
1996
1997                                 intel_pt_set_ip(decoder);
1998                                 if (current_ip)
1999                                         intel_pt_log_to("Setting IP",
2000                                                         decoder->ip);
2001                         }
2002                         break;
2003
2004                 case INTEL_PT_MTC:
2005                         intel_pt_calc_mtc_timestamp(decoder);
2006                         break;
2007
2008                 case INTEL_PT_TSC:
2009                         intel_pt_calc_tsc_timestamp(decoder);
2010                         break;
2011
2012                 case INTEL_PT_TMA:
2013                         intel_pt_calc_tma(decoder);
2014                         break;
2015
2016                 case INTEL_PT_CYC:
2017                         intel_pt_calc_cyc_timestamp(decoder);
2018                         break;
2019
2020                 case INTEL_PT_CBR:
2021                         intel_pt_calc_cbr(decoder);
2022                         break;
2023
2024                 case INTEL_PT_PIP:
2025                         decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
2026                         break;
2027
2028                 case INTEL_PT_MODE_EXEC:
2029                         decoder->exec_mode = decoder->packet.payload;
2030                         break;
2031
2032                 case INTEL_PT_MODE_TSX:
2033                         intel_pt_update_in_tx(decoder);
2034                         break;
2035
2036                 case INTEL_PT_TRACESTOP:
2037                         decoder->pge = false;
2038                         decoder->continuous_period = false;
2039                         intel_pt_clear_tx_flags(decoder);
2040                         __fallthrough;
2041
2042                 case INTEL_PT_TNT:
2043                         decoder->have_tma = false;
2044                         intel_pt_log("ERROR: Unexpected packet\n");
2045                         if (decoder->ip)
2046                                 decoder->pkt_state = INTEL_PT_STATE_ERR4;
2047                         else
2048                                 decoder->pkt_state = INTEL_PT_STATE_ERR3;
2049                         return -ENOENT;
2050
2051                 case INTEL_PT_BAD: /* Does not happen */
2052                         return intel_pt_bug(decoder);
2053
2054                 case INTEL_PT_OVF:
2055                         return intel_pt_overflow(decoder);
2056
2057                 case INTEL_PT_PSBEND:
2058                         return 0;
2059
2060                 case INTEL_PT_PSB:
2061                 case INTEL_PT_VMCS:
2062                 case INTEL_PT_MNT:
2063                 case INTEL_PT_PAD:
2064                 default:
2065                         break;
2066                 }
2067         }
2068 }
2069
2070 static int intel_pt_walk_to_ip(struct intel_pt_decoder *decoder)
2071 {
2072         int err;
2073
2074         while (1) {
2075                 err = intel_pt_get_next_packet(decoder);
2076                 if (err)
2077                         return err;
2078
2079                 switch (decoder->packet.type) {
2080                 case INTEL_PT_TIP_PGD:
2081                         decoder->continuous_period = false;
2082                         __fallthrough;
2083                 case INTEL_PT_TIP_PGE:
2084                 case INTEL_PT_TIP:
2085                         decoder->pge = decoder->packet.type != INTEL_PT_TIP_PGD;
2086                         if (intel_pt_have_ip(decoder))
2087                                 intel_pt_set_ip(decoder);
2088                         if (!decoder->ip)
2089                                 break;
2090                         if (decoder->packet.type == INTEL_PT_TIP_PGE)
2091                                 decoder->state.type |= INTEL_PT_TRACE_BEGIN;
2092                         if (decoder->packet.type == INTEL_PT_TIP_PGD)
2093                                 decoder->state.type |= INTEL_PT_TRACE_END;
2094                         return 0;
2095
2096                 case INTEL_PT_FUP:
2097                         if (intel_pt_have_ip(decoder))
2098                                 intel_pt_set_ip(decoder);
2099                         if (decoder->ip)
2100                                 return 0;
2101                         break;
2102
2103                 case INTEL_PT_MTC:
2104                         intel_pt_calc_mtc_timestamp(decoder);
2105                         break;
2106
2107                 case INTEL_PT_TSC:
2108                         intel_pt_calc_tsc_timestamp(decoder);
2109                         break;
2110
2111                 case INTEL_PT_TMA:
2112                         intel_pt_calc_tma(decoder);
2113                         break;
2114
2115                 case INTEL_PT_CYC:
2116                         intel_pt_calc_cyc_timestamp(decoder);
2117                         break;
2118
2119                 case INTEL_PT_CBR:
2120                         intel_pt_calc_cbr(decoder);
2121                         break;
2122
2123                 case INTEL_PT_PIP:
2124                         decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
2125                         break;
2126
2127                 case INTEL_PT_MODE_EXEC:
2128                         decoder->exec_mode = decoder->packet.payload;
2129                         break;
2130
2131                 case INTEL_PT_MODE_TSX:
2132                         intel_pt_update_in_tx(decoder);
2133                         break;
2134
2135                 case INTEL_PT_OVF:
2136                         return intel_pt_overflow(decoder);
2137
2138                 case INTEL_PT_BAD: /* Does not happen */
2139                         return intel_pt_bug(decoder);
2140
2141                 case INTEL_PT_TRACESTOP:
2142                         decoder->pge = false;
2143                         decoder->continuous_period = false;
2144                         intel_pt_clear_tx_flags(decoder);
2145                         decoder->have_tma = false;
2146                         break;
2147
2148                 case INTEL_PT_PSB:
2149                         decoder->last_ip = 0;
2150                         decoder->have_last_ip = true;
2151                         intel_pt_clear_stack(&decoder->stack);
2152                         err = intel_pt_walk_psb(decoder);
2153                         if (err)
2154                                 return err;
2155                         if (decoder->ip) {
2156                                 /* Do not have a sample */
2157                                 decoder->state.type = 0;
2158                                 return 0;
2159                         }
2160                         break;
2161
2162                 case INTEL_PT_TNT:
2163                 case INTEL_PT_PSBEND:
2164                 case INTEL_PT_VMCS:
2165                 case INTEL_PT_MNT:
2166                 case INTEL_PT_PAD:
2167                 case INTEL_PT_PTWRITE:
2168                 case INTEL_PT_PTWRITE_IP:
2169                 case INTEL_PT_EXSTOP:
2170                 case INTEL_PT_EXSTOP_IP:
2171                 case INTEL_PT_MWAIT:
2172                 case INTEL_PT_PWRE:
2173                 case INTEL_PT_PWRX:
2174                 default:
2175                         break;
2176                 }
2177         }
2178 }
2179
2180 static int intel_pt_sync_ip(struct intel_pt_decoder *decoder)
2181 {
2182         int err;
2183
2184         decoder->set_fup_tx_flags = false;
2185         decoder->set_fup_ptw = false;
2186         decoder->set_fup_mwait = false;
2187         decoder->set_fup_pwre = false;
2188         decoder->set_fup_exstop = false;
2189
2190         if (!decoder->branch_enable) {
2191                 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2192                 decoder->overflow = false;
2193                 decoder->state.type = 0; /* Do not have a sample */
2194                 return 0;
2195         }
2196
2197         intel_pt_log("Scanning for full IP\n");
2198         err = intel_pt_walk_to_ip(decoder);
2199         if (err)
2200                 return err;
2201
2202         decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2203         decoder->overflow = false;
2204
2205         decoder->state.from_ip = 0;
2206         decoder->state.to_ip = decoder->ip;
2207         intel_pt_log_to("Setting IP", decoder->ip);
2208
2209         return 0;
2210 }
2211
2212 static int intel_pt_part_psb(struct intel_pt_decoder *decoder)
2213 {
2214         const unsigned char *end = decoder->buf + decoder->len;
2215         size_t i;
2216
2217         for (i = INTEL_PT_PSB_LEN - 1; i; i--) {
2218                 if (i > decoder->len)
2219                         continue;
2220                 if (!memcmp(end - i, INTEL_PT_PSB_STR, i))
2221                         return i;
2222         }
2223         return 0;
2224 }
2225
2226 static int intel_pt_rest_psb(struct intel_pt_decoder *decoder, int part_psb)
2227 {
2228         size_t rest_psb = INTEL_PT_PSB_LEN - part_psb;
2229         const char *psb = INTEL_PT_PSB_STR;
2230
2231         if (rest_psb > decoder->len ||
2232             memcmp(decoder->buf, psb + part_psb, rest_psb))
2233                 return 0;
2234
2235         return rest_psb;
2236 }
2237
2238 static int intel_pt_get_split_psb(struct intel_pt_decoder *decoder,
2239                                   int part_psb)
2240 {
2241         int rest_psb, ret;
2242
2243         decoder->pos += decoder->len;
2244         decoder->len = 0;
2245
2246         ret = intel_pt_get_next_data(decoder);
2247         if (ret)
2248                 return ret;
2249
2250         rest_psb = intel_pt_rest_psb(decoder, part_psb);
2251         if (!rest_psb)
2252                 return 0;
2253
2254         decoder->pos -= part_psb;
2255         decoder->next_buf = decoder->buf + rest_psb;
2256         decoder->next_len = decoder->len - rest_psb;
2257         memcpy(decoder->temp_buf, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2258         decoder->buf = decoder->temp_buf;
2259         decoder->len = INTEL_PT_PSB_LEN;
2260
2261         return 0;
2262 }
2263
2264 static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder)
2265 {
2266         unsigned char *next;
2267         int ret;
2268
2269         intel_pt_log("Scanning for PSB\n");
2270         while (1) {
2271                 if (!decoder->len) {
2272                         ret = intel_pt_get_next_data(decoder);
2273                         if (ret)
2274                                 return ret;
2275                 }
2276
2277                 next = memmem(decoder->buf, decoder->len, INTEL_PT_PSB_STR,
2278                               INTEL_PT_PSB_LEN);
2279                 if (!next) {
2280                         int part_psb;
2281
2282                         part_psb = intel_pt_part_psb(decoder);
2283                         if (part_psb) {
2284                                 ret = intel_pt_get_split_psb(decoder, part_psb);
2285                                 if (ret)
2286                                         return ret;
2287                         } else {
2288                                 decoder->pos += decoder->len;
2289                                 decoder->len = 0;
2290                         }
2291                         continue;
2292                 }
2293
2294                 decoder->pkt_step = next - decoder->buf;
2295                 return intel_pt_get_next_packet(decoder);
2296         }
2297 }
2298
2299 static int intel_pt_sync(struct intel_pt_decoder *decoder)
2300 {
2301         int err;
2302
2303         decoder->pge = false;
2304         decoder->continuous_period = false;
2305         decoder->have_last_ip = false;
2306         decoder->last_ip = 0;
2307         decoder->ip = 0;
2308         intel_pt_clear_stack(&decoder->stack);
2309
2310         err = intel_pt_scan_for_psb(decoder);
2311         if (err)
2312                 return err;
2313
2314         decoder->have_last_ip = true;
2315         decoder->pkt_state = INTEL_PT_STATE_NO_IP;
2316
2317         err = intel_pt_walk_psb(decoder);
2318         if (err)
2319                 return err;
2320
2321         if (decoder->ip) {
2322                 decoder->state.type = 0; /* Do not have a sample */
2323                 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2324         } else {
2325                 return intel_pt_sync_ip(decoder);
2326         }
2327
2328         return 0;
2329 }
2330
2331 static uint64_t intel_pt_est_timestamp(struct intel_pt_decoder *decoder)
2332 {
2333         uint64_t est = decoder->sample_insn_cnt << 1;
2334
2335         if (!decoder->cbr || !decoder->max_non_turbo_ratio)
2336                 goto out;
2337
2338         est *= decoder->max_non_turbo_ratio;
2339         est /= decoder->cbr;
2340 out:
2341         return decoder->sample_timestamp + est;
2342 }
2343
2344 const struct intel_pt_state *intel_pt_decode(struct intel_pt_decoder *decoder)
2345 {
2346         int err;
2347
2348         do {
2349                 decoder->state.type = INTEL_PT_BRANCH;
2350                 decoder->state.flags = 0;
2351
2352                 switch (decoder->pkt_state) {
2353                 case INTEL_PT_STATE_NO_PSB:
2354                         err = intel_pt_sync(decoder);
2355                         break;
2356                 case INTEL_PT_STATE_NO_IP:
2357                         decoder->have_last_ip = false;
2358                         decoder->last_ip = 0;
2359                         decoder->ip = 0;
2360                         __fallthrough;
2361                 case INTEL_PT_STATE_ERR_RESYNC:
2362                         err = intel_pt_sync_ip(decoder);
2363                         break;
2364                 case INTEL_PT_STATE_IN_SYNC:
2365                         err = intel_pt_walk_trace(decoder);
2366                         break;
2367                 case INTEL_PT_STATE_TNT:
2368                         err = intel_pt_walk_tnt(decoder);
2369                         if (err == -EAGAIN)
2370                                 err = intel_pt_walk_trace(decoder);
2371                         break;
2372                 case INTEL_PT_STATE_TIP:
2373                 case INTEL_PT_STATE_TIP_PGD:
2374                         err = intel_pt_walk_tip(decoder);
2375                         break;
2376                 case INTEL_PT_STATE_FUP:
2377                         decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2378                         err = intel_pt_walk_fup(decoder);
2379                         if (err == -EAGAIN)
2380                                 err = intel_pt_walk_fup_tip(decoder);
2381                         else if (!err)
2382                                 decoder->pkt_state = INTEL_PT_STATE_FUP;
2383                         break;
2384                 case INTEL_PT_STATE_FUP_NO_TIP:
2385                         decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2386                         err = intel_pt_walk_fup(decoder);
2387                         if (err == -EAGAIN)
2388                                 err = intel_pt_walk_trace(decoder);
2389                         break;
2390                 default:
2391                         err = intel_pt_bug(decoder);
2392                         break;
2393                 }
2394         } while (err == -ENOLINK);
2395
2396         if (err) {
2397                 decoder->state.err = intel_pt_ext_err(err);
2398                 decoder->state.from_ip = decoder->ip;
2399                 decoder->sample_timestamp = decoder->timestamp;
2400                 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
2401         } else {
2402                 decoder->state.err = 0;
2403                 if (decoder->cbr != decoder->cbr_seen && decoder->state.type) {
2404                         decoder->cbr_seen = decoder->cbr;
2405                         decoder->state.type |= INTEL_PT_CBR_CHG;
2406                         decoder->state.cbr_payload = decoder->cbr_payload;
2407                 }
2408                 if (intel_pt_sample_time(decoder->pkt_state)) {
2409                         decoder->sample_timestamp = decoder->timestamp;
2410                         decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
2411                 }
2412         }
2413
2414         decoder->state.timestamp = decoder->sample_timestamp;
2415         decoder->state.est_timestamp = intel_pt_est_timestamp(decoder);
2416         decoder->state.cr3 = decoder->cr3;
2417         decoder->state.tot_insn_cnt = decoder->tot_insn_cnt;
2418
2419         return &decoder->state;
2420 }
2421
2422 /**
2423  * intel_pt_next_psb - move buffer pointer to the start of the next PSB packet.
2424  * @buf: pointer to buffer pointer
2425  * @len: size of buffer
2426  *
2427  * Updates the buffer pointer to point to the start of the next PSB packet if
2428  * there is one, otherwise the buffer pointer is unchanged.  If @buf is updated,
2429  * @len is adjusted accordingly.
2430  *
2431  * Return: %true if a PSB packet is found, %false otherwise.
2432  */
2433 static bool intel_pt_next_psb(unsigned char **buf, size_t *len)
2434 {
2435         unsigned char *next;
2436
2437         next = memmem(*buf, *len, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2438         if (next) {
2439                 *len -= next - *buf;
2440                 *buf = next;
2441                 return true;
2442         }
2443         return false;
2444 }
2445
2446 /**
2447  * intel_pt_step_psb - move buffer pointer to the start of the following PSB
2448  *                     packet.
2449  * @buf: pointer to buffer pointer
2450  * @len: size of buffer
2451  *
2452  * Updates the buffer pointer to point to the start of the following PSB packet
2453  * (skipping the PSB at @buf itself) if there is one, otherwise the buffer
2454  * pointer is unchanged.  If @buf is updated, @len is adjusted accordingly.
2455  *
2456  * Return: %true if a PSB packet is found, %false otherwise.
2457  */
2458 static bool intel_pt_step_psb(unsigned char **buf, size_t *len)
2459 {
2460         unsigned char *next;
2461
2462         if (!*len)
2463                 return false;
2464
2465         next = memmem(*buf + 1, *len - 1, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2466         if (next) {
2467                 *len -= next - *buf;
2468                 *buf = next;
2469                 return true;
2470         }
2471         return false;
2472 }
2473
2474 /**
2475  * intel_pt_last_psb - find the last PSB packet in a buffer.
2476  * @buf: buffer
2477  * @len: size of buffer
2478  *
2479  * This function finds the last PSB in a buffer.
2480  *
2481  * Return: A pointer to the last PSB in @buf if found, %NULL otherwise.
2482  */
2483 static unsigned char *intel_pt_last_psb(unsigned char *buf, size_t len)
2484 {
2485         const char *n = INTEL_PT_PSB_STR;
2486         unsigned char *p;
2487         size_t k;
2488
2489         if (len < INTEL_PT_PSB_LEN)
2490                 return NULL;
2491
2492         k = len - INTEL_PT_PSB_LEN + 1;
2493         while (1) {
2494                 p = memrchr(buf, n[0], k);
2495                 if (!p)
2496                         return NULL;
2497                 if (!memcmp(p + 1, n + 1, INTEL_PT_PSB_LEN - 1))
2498                         return p;
2499                 k = p - buf;
2500                 if (!k)
2501                         return NULL;
2502         }
2503 }
2504
2505 /**
2506  * intel_pt_next_tsc - find and return next TSC.
2507  * @buf: buffer
2508  * @len: size of buffer
2509  * @tsc: TSC value returned
2510  * @rem: returns remaining size when TSC is found
2511  *
2512  * Find a TSC packet in @buf and return the TSC value.  This function assumes
2513  * that @buf starts at a PSB and that PSB+ will contain TSC and so stops if a
2514  * PSBEND packet is found.
2515  *
2516  * Return: %true if TSC is found, false otherwise.
2517  */
2518 static bool intel_pt_next_tsc(unsigned char *buf, size_t len, uint64_t *tsc,
2519                               size_t *rem)
2520 {
2521         struct intel_pt_pkt packet;
2522         int ret;
2523
2524         while (len) {
2525                 ret = intel_pt_get_packet(buf, len, &packet);
2526                 if (ret <= 0)
2527                         return false;
2528                 if (packet.type == INTEL_PT_TSC) {
2529                         *tsc = packet.payload;
2530                         *rem = len;
2531                         return true;
2532                 }
2533                 if (packet.type == INTEL_PT_PSBEND)
2534                         return false;
2535                 buf += ret;
2536                 len -= ret;
2537         }
2538         return false;
2539 }
2540
2541 /**
2542  * intel_pt_tsc_cmp - compare 7-byte TSCs.
2543  * @tsc1: first TSC to compare
2544  * @tsc2: second TSC to compare
2545  *
2546  * This function compares 7-byte TSC values allowing for the possibility that
2547  * TSC wrapped around.  Generally it is not possible to know if TSC has wrapped
2548  * around so for that purpose this function assumes the absolute difference is
2549  * less than half the maximum difference.
2550  *
2551  * Return: %-1 if @tsc1 is before @tsc2, %0 if @tsc1 == @tsc2, %1 if @tsc1 is
2552  * after @tsc2.
2553  */
2554 static int intel_pt_tsc_cmp(uint64_t tsc1, uint64_t tsc2)
2555 {
2556         const uint64_t halfway = (1ULL << 55);
2557
2558         if (tsc1 == tsc2)
2559                 return 0;
2560
2561         if (tsc1 < tsc2) {
2562                 if (tsc2 - tsc1 < halfway)
2563                         return -1;
2564                 else
2565                         return 1;
2566         } else {
2567                 if (tsc1 - tsc2 < halfway)
2568                         return 1;
2569                 else
2570                         return -1;
2571         }
2572 }
2573
2574 /**
2575  * intel_pt_find_overlap_tsc - determine start of non-overlapped trace data
2576  *                             using TSC.
2577  * @buf_a: first buffer
2578  * @len_a: size of first buffer
2579  * @buf_b: second buffer
2580  * @len_b: size of second buffer
2581  * @consecutive: returns true if there is data in buf_b that is consecutive
2582  *               to buf_a
2583  *
2584  * If the trace contains TSC we can look at the last TSC of @buf_a and the
2585  * first TSC of @buf_b in order to determine if the buffers overlap, and then
2586  * walk forward in @buf_b until a later TSC is found.  A precondition is that
2587  * @buf_a and @buf_b are positioned at a PSB.
2588  *
2589  * Return: A pointer into @buf_b from where non-overlapped data starts, or
2590  * @buf_b + @len_b if there is no non-overlapped data.
2591  */
2592 static unsigned char *intel_pt_find_overlap_tsc(unsigned char *buf_a,
2593                                                 size_t len_a,
2594                                                 unsigned char *buf_b,
2595                                                 size_t len_b, bool *consecutive)
2596 {
2597         uint64_t tsc_a, tsc_b;
2598         unsigned char *p;
2599         size_t len, rem_a, rem_b;
2600
2601         p = intel_pt_last_psb(buf_a, len_a);
2602         if (!p)
2603                 return buf_b; /* No PSB in buf_a => no overlap */
2604
2605         len = len_a - (p - buf_a);
2606         if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a)) {
2607                 /* The last PSB+ in buf_a is incomplete, so go back one more */
2608                 len_a -= len;
2609                 p = intel_pt_last_psb(buf_a, len_a);
2610                 if (!p)
2611                         return buf_b; /* No full PSB+ => assume no overlap */
2612                 len = len_a - (p - buf_a);
2613                 if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a))
2614                         return buf_b; /* No TSC in buf_a => assume no overlap */
2615         }
2616
2617         while (1) {
2618                 /* Ignore PSB+ with no TSC */
2619                 if (intel_pt_next_tsc(buf_b, len_b, &tsc_b, &rem_b)) {
2620                         int cmp = intel_pt_tsc_cmp(tsc_a, tsc_b);
2621
2622                         /* Same TSC, so buffers are consecutive */
2623                         if (!cmp && rem_b >= rem_a) {
2624                                 *consecutive = true;
2625                                 return buf_b + len_b - (rem_b - rem_a);
2626                         }
2627                         if (cmp < 0)
2628                                 return buf_b; /* tsc_a < tsc_b => no overlap */
2629                 }
2630
2631                 if (!intel_pt_step_psb(&buf_b, &len_b))
2632                         return buf_b + len_b; /* No PSB in buf_b => no data */
2633         }
2634 }
2635
2636 /**
2637  * intel_pt_find_overlap - determine start of non-overlapped trace data.
2638  * @buf_a: first buffer
2639  * @len_a: size of first buffer
2640  * @buf_b: second buffer
2641  * @len_b: size of second buffer
2642  * @have_tsc: can use TSC packets to detect overlap
2643  * @consecutive: returns true if there is data in buf_b that is consecutive
2644  *               to buf_a
2645  *
2646  * When trace samples or snapshots are recorded there is the possibility that
2647  * the data overlaps.  Note that, for the purposes of decoding, data is only
2648  * useful if it begins with a PSB packet.
2649  *
2650  * Return: A pointer into @buf_b from where non-overlapped data starts, or
2651  * @buf_b + @len_b if there is no non-overlapped data.
2652  */
2653 unsigned char *intel_pt_find_overlap(unsigned char *buf_a, size_t len_a,
2654                                      unsigned char *buf_b, size_t len_b,
2655                                      bool have_tsc, bool *consecutive)
2656 {
2657         unsigned char *found;
2658
2659         /* Buffer 'b' must start at PSB so throw away everything before that */
2660         if (!intel_pt_next_psb(&buf_b, &len_b))
2661                 return buf_b + len_b; /* No PSB */
2662
2663         if (!intel_pt_next_psb(&buf_a, &len_a))
2664                 return buf_b; /* No overlap */
2665
2666         if (have_tsc) {
2667                 found = intel_pt_find_overlap_tsc(buf_a, len_a, buf_b, len_b,
2668                                                   consecutive);
2669                 if (found)
2670                         return found;
2671         }
2672
2673         /*
2674          * Buffer 'b' cannot end within buffer 'a' so, for comparison purposes,
2675          * we can ignore the first part of buffer 'a'.
2676          */
2677         while (len_b < len_a) {
2678                 if (!intel_pt_step_psb(&buf_a, &len_a))
2679                         return buf_b; /* No overlap */
2680         }
2681
2682         /* Now len_b >= len_a */
2683         while (1) {
2684                 /* Potential overlap so check the bytes */
2685                 found = memmem(buf_a, len_a, buf_b, len_a);
2686                 if (found) {
2687                         *consecutive = true;
2688                         return buf_b + len_a;
2689                 }
2690
2691                 /* Try again at next PSB in buffer 'a' */
2692                 if (!intel_pt_step_psb(&buf_a, &len_a))
2693                         return buf_b; /* No overlap */
2694         }
2695 }