2 * intel_pt_decoder.c: Intel Processor Trace support
3 * Copyright (c) 2013-2014, Intel Corporation.
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.
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
25 #include <linux/compiler.h>
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"
35 #define INTEL_PT_BLK_SIZE 1024
37 #define BIT63 (((uint64_t)1 << 63))
39 #define INTEL_PT_RETURN 1
41 /* Maximum number of loops with no packets consumed i.e. stuck in a loop */
42 #define INTEL_PT_MAX_LOOPS 10000
45 struct intel_pt_blk *prev;
46 uint64_t ip[INTEL_PT_BLK_SIZE];
49 struct intel_pt_stack {
50 struct intel_pt_blk *blk;
51 struct intel_pt_blk *spare;
55 enum intel_pt_pkt_state {
56 INTEL_PT_STATE_NO_PSB,
58 INTEL_PT_STATE_ERR_RESYNC,
59 INTEL_PT_STATE_IN_SYNC,
62 INTEL_PT_STATE_TIP_PGD,
64 INTEL_PT_STATE_FUP_NO_TIP,
67 static inline bool intel_pt_sample_time(enum intel_pt_pkt_state 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:
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:
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
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
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);
105 struct intel_pt_state state;
106 const unsigned char *buf;
108 bool return_compression;
121 uint64_t tsc_timestamp;
122 uint64_t ref_timestamp;
123 uint64_t sample_timestamp;
125 uint64_t ctc_timestamp;
128 uint64_t cyc_ref_timestamp;
130 uint32_t tsc_ctc_ratio_n;
131 uint32_t tsc_ctc_ratio_d;
132 uint32_t tsc_ctc_mult;
134 uint32_t ctc_rem_mask;
136 struct intel_pt_stack stack;
137 enum intel_pt_pkt_state pkt_state;
138 struct intel_pt_pkt packet;
139 struct intel_pt_pkt tnt;
142 int last_packet_type;
144 unsigned int cbr_seen;
145 unsigned int max_non_turbo_ratio;
146 double max_non_turbo_ratio_fp;
147 double cbr_cyc_to_tsc;
148 double calc_cyc_to_tsc;
149 bool have_calc_cyc_to_tsc;
151 unsigned int insn_bytes;
153 enum intel_pt_period_type period_type;
154 uint64_t tot_insn_cnt;
155 uint64_t period_insn_cnt;
156 uint64_t period_mask;
157 uint64_t period_ticks;
158 uint64_t last_masked_timestamp;
159 bool continuous_period;
161 bool set_fup_tx_flags;
166 unsigned int fup_tx_flags;
167 unsigned int tx_flags;
168 uint64_t fup_ptw_payload;
169 uint64_t fup_mwait_payload;
170 uint64_t fup_pwre_payload;
171 uint64_t cbr_payload;
172 uint64_t timestamp_insn_cnt;
173 uint64_t sample_insn_cnt;
178 const unsigned char *next_buf;
180 unsigned char temp_buf[INTEL_PT_PKT_MAX_SZ];
183 static uint64_t intel_pt_lower_power_of_2(uint64_t x)
187 for (i = 0; x != 1; i++)
193 static void intel_pt_setup_period(struct intel_pt_decoder *decoder)
195 if (decoder->period_type == INTEL_PT_PERIOD_TICKS) {
198 period = intel_pt_lower_power_of_2(decoder->period);
199 decoder->period_mask = ~(period - 1);
200 decoder->period_ticks = period;
204 static uint64_t multdiv(uint64_t t, uint32_t n, uint32_t d)
208 return (t / d) * n + ((t % d) * n) / d;
211 struct intel_pt_decoder *intel_pt_decoder_new(struct intel_pt_params *params)
213 struct intel_pt_decoder *decoder;
215 if (!params->get_trace || !params->walk_insn)
218 decoder = zalloc(sizeof(struct intel_pt_decoder));
222 decoder->get_trace = params->get_trace;
223 decoder->walk_insn = params->walk_insn;
224 decoder->pgd_ip = params->pgd_ip;
225 decoder->data = params->data;
226 decoder->return_compression = params->return_compression;
227 decoder->branch_enable = params->branch_enable;
229 decoder->period = params->period;
230 decoder->period_type = params->period_type;
232 decoder->max_non_turbo_ratio = params->max_non_turbo_ratio;
233 decoder->max_non_turbo_ratio_fp = params->max_non_turbo_ratio;
235 intel_pt_setup_period(decoder);
237 decoder->mtc_shift = params->mtc_period;
238 decoder->ctc_rem_mask = (1 << decoder->mtc_shift) - 1;
240 decoder->tsc_ctc_ratio_n = params->tsc_ctc_ratio_n;
241 decoder->tsc_ctc_ratio_d = params->tsc_ctc_ratio_d;
243 if (!decoder->tsc_ctc_ratio_n)
244 decoder->tsc_ctc_ratio_d = 0;
246 if (decoder->tsc_ctc_ratio_d) {
247 if (!(decoder->tsc_ctc_ratio_n % decoder->tsc_ctc_ratio_d))
248 decoder->tsc_ctc_mult = decoder->tsc_ctc_ratio_n /
249 decoder->tsc_ctc_ratio_d;
252 * Allow for timestamps appearing to backwards because a TSC
253 * packet has slipped past a MTC packet, so allow 2 MTC ticks
256 decoder->tsc_slip = multdiv(2 << decoder->mtc_shift,
257 decoder->tsc_ctc_ratio_n,
258 decoder->tsc_ctc_ratio_d);
260 /* ... or 0x100 paranoia */
261 if (decoder->tsc_slip < 0x100)
262 decoder->tsc_slip = 0x100;
264 intel_pt_log("timestamp: mtc_shift %u\n", decoder->mtc_shift);
265 intel_pt_log("timestamp: tsc_ctc_ratio_n %u\n", decoder->tsc_ctc_ratio_n);
266 intel_pt_log("timestamp: tsc_ctc_ratio_d %u\n", decoder->tsc_ctc_ratio_d);
267 intel_pt_log("timestamp: tsc_ctc_mult %u\n", decoder->tsc_ctc_mult);
268 intel_pt_log("timestamp: tsc_slip %#x\n", decoder->tsc_slip);
273 static void intel_pt_pop_blk(struct intel_pt_stack *stack)
275 struct intel_pt_blk *blk = stack->blk;
277 stack->blk = blk->prev;
284 static uint64_t intel_pt_pop(struct intel_pt_stack *stack)
289 intel_pt_pop_blk(stack);
292 stack->pos = INTEL_PT_BLK_SIZE;
294 return stack->blk->ip[--stack->pos];
297 static int intel_pt_alloc_blk(struct intel_pt_stack *stack)
299 struct intel_pt_blk *blk;
305 blk = malloc(sizeof(struct intel_pt_blk));
310 blk->prev = stack->blk;
316 static int intel_pt_push(struct intel_pt_stack *stack, uint64_t ip)
320 if (!stack->blk || stack->pos == INTEL_PT_BLK_SIZE) {
321 err = intel_pt_alloc_blk(stack);
326 stack->blk->ip[stack->pos++] = ip;
330 static void intel_pt_clear_stack(struct intel_pt_stack *stack)
333 intel_pt_pop_blk(stack);
337 static void intel_pt_free_stack(struct intel_pt_stack *stack)
339 intel_pt_clear_stack(stack);
341 zfree(&stack->spare);
344 void intel_pt_decoder_free(struct intel_pt_decoder *decoder)
346 intel_pt_free_stack(&decoder->stack);
350 static int intel_pt_ext_err(int code)
354 return INTEL_PT_ERR_NOMEM;
356 return INTEL_PT_ERR_INTERN;
358 return INTEL_PT_ERR_BADPKT;
360 return INTEL_PT_ERR_NODATA;
362 return INTEL_PT_ERR_NOINSN;
364 return INTEL_PT_ERR_MISMAT;
366 return INTEL_PT_ERR_OVR;
368 return INTEL_PT_ERR_LOST;
370 return INTEL_PT_ERR_NELOOP;
372 return INTEL_PT_ERR_UNK;
376 static const char *intel_pt_err_msgs[] = {
377 [INTEL_PT_ERR_NOMEM] = "Memory allocation failed",
378 [INTEL_PT_ERR_INTERN] = "Internal error",
379 [INTEL_PT_ERR_BADPKT] = "Bad packet",
380 [INTEL_PT_ERR_NODATA] = "No more data",
381 [INTEL_PT_ERR_NOINSN] = "Failed to get instruction",
382 [INTEL_PT_ERR_MISMAT] = "Trace doesn't match instruction",
383 [INTEL_PT_ERR_OVR] = "Overflow packet",
384 [INTEL_PT_ERR_LOST] = "Lost trace data",
385 [INTEL_PT_ERR_UNK] = "Unknown error!",
386 [INTEL_PT_ERR_NELOOP] = "Never-ending loop",
389 int intel_pt__strerror(int code, char *buf, size_t buflen)
391 if (code < 1 || code >= INTEL_PT_ERR_MAX)
392 code = INTEL_PT_ERR_UNK;
393 strlcpy(buf, intel_pt_err_msgs[code], buflen);
397 static uint64_t intel_pt_calc_ip(const struct intel_pt_pkt *packet,
402 switch (packet->count) {
404 ip = (last_ip & (uint64_t)0xffffffffffff0000ULL) |
408 ip = (last_ip & (uint64_t)0xffffffff00000000ULL) |
412 ip = packet->payload;
413 /* Sign-extend 6-byte ip */
414 if (ip & (uint64_t)0x800000000000ULL)
415 ip |= (uint64_t)0xffff000000000000ULL;
418 ip = (last_ip & (uint64_t)0xffff000000000000ULL) |
422 ip = packet->payload;
431 static inline void intel_pt_set_last_ip(struct intel_pt_decoder *decoder)
433 decoder->last_ip = intel_pt_calc_ip(&decoder->packet, decoder->last_ip);
434 decoder->have_last_ip = true;
437 static inline void intel_pt_set_ip(struct intel_pt_decoder *decoder)
439 intel_pt_set_last_ip(decoder);
440 decoder->ip = decoder->last_ip;
443 static void intel_pt_decoder_log_packet(struct intel_pt_decoder *decoder)
445 intel_pt_log_packet(&decoder->packet, decoder->pkt_len, decoder->pos,
449 static int intel_pt_bug(struct intel_pt_decoder *decoder)
451 intel_pt_log("ERROR: Internal error\n");
452 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
456 static inline void intel_pt_clear_tx_flags(struct intel_pt_decoder *decoder)
458 decoder->tx_flags = 0;
461 static inline void intel_pt_update_in_tx(struct intel_pt_decoder *decoder)
463 decoder->tx_flags = decoder->packet.payload & INTEL_PT_IN_TX;
466 static int intel_pt_bad_packet(struct intel_pt_decoder *decoder)
468 intel_pt_clear_tx_flags(decoder);
469 decoder->have_tma = false;
470 decoder->pkt_len = 1;
471 decoder->pkt_step = 1;
472 intel_pt_decoder_log_packet(decoder);
473 if (decoder->pkt_state != INTEL_PT_STATE_NO_PSB) {
474 intel_pt_log("ERROR: Bad packet\n");
475 decoder->pkt_state = INTEL_PT_STATE_ERR1;
480 static int intel_pt_get_data(struct intel_pt_decoder *decoder)
482 struct intel_pt_buffer buffer = { .buf = 0, };
485 decoder->pkt_step = 0;
487 intel_pt_log("Getting more data\n");
488 ret = decoder->get_trace(&buffer, decoder->data);
491 decoder->buf = buffer.buf;
492 decoder->len = buffer.len;
494 intel_pt_log("No more data\n");
497 if (!buffer.consecutive) {
499 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
500 decoder->ref_timestamp = buffer.ref_timestamp;
501 decoder->timestamp = 0;
502 decoder->have_tma = false;
503 decoder->state.trace_nr = buffer.trace_nr;
504 intel_pt_log("Reference timestamp 0x%" PRIx64 "\n",
505 decoder->ref_timestamp);
512 static int intel_pt_get_next_data(struct intel_pt_decoder *decoder)
514 if (!decoder->next_buf)
515 return intel_pt_get_data(decoder);
517 decoder->buf = decoder->next_buf;
518 decoder->len = decoder->next_len;
519 decoder->next_buf = 0;
520 decoder->next_len = 0;
524 static int intel_pt_get_split_packet(struct intel_pt_decoder *decoder)
526 unsigned char *buf = decoder->temp_buf;
527 size_t old_len, len, n;
530 old_len = decoder->len;
532 memcpy(buf, decoder->buf, len);
534 ret = intel_pt_get_data(decoder);
536 decoder->pos += old_len;
537 return ret < 0 ? ret : -EINVAL;
540 n = INTEL_PT_PKT_MAX_SZ - len;
541 if (n > decoder->len)
543 memcpy(buf + len, decoder->buf, n);
546 ret = intel_pt_get_packet(buf, len, &decoder->packet);
547 if (ret < (int)old_len) {
548 decoder->next_buf = decoder->buf;
549 decoder->next_len = decoder->len;
551 decoder->len = old_len;
552 return intel_pt_bad_packet(decoder);
555 decoder->next_buf = decoder->buf + (ret - old_len);
556 decoder->next_len = decoder->len - (ret - old_len);
564 struct intel_pt_pkt_info {
565 struct intel_pt_decoder *decoder;
566 struct intel_pt_pkt packet;
569 int last_packet_type;
573 typedef int (*intel_pt_pkt_cb_t)(struct intel_pt_pkt_info *pkt_info);
575 /* Lookahead packets in current buffer */
576 static int intel_pt_pkt_lookahead(struct intel_pt_decoder *decoder,
577 intel_pt_pkt_cb_t cb, void *data)
579 struct intel_pt_pkt_info pkt_info;
580 const unsigned char *buf = decoder->buf;
581 size_t len = decoder->len;
584 pkt_info.decoder = decoder;
585 pkt_info.pos = decoder->pos;
586 pkt_info.pkt_len = decoder->pkt_step;
587 pkt_info.last_packet_type = decoder->last_packet_type;
588 pkt_info.data = data;
592 pkt_info.pos += pkt_info.pkt_len;
593 buf += pkt_info.pkt_len;
594 len -= pkt_info.pkt_len;
597 return INTEL_PT_NEED_MORE_BYTES;
599 ret = intel_pt_get_packet(buf, len, &pkt_info.packet);
601 return INTEL_PT_NEED_MORE_BYTES;
605 pkt_info.pkt_len = ret;
606 } while (pkt_info.packet.type == INTEL_PT_PAD);
612 pkt_info.last_packet_type = pkt_info.packet.type;
616 struct intel_pt_calc_cyc_to_tsc_info {
620 uint64_t ctc_timestamp;
622 uint64_t tsc_timestamp;
627 double cbr_cyc_to_tsc;
631 * MTC provides a 8-bit slice of CTC but the TMA packet only provides the lower
632 * 16 bits of CTC. If mtc_shift > 8 then some of the MTC bits are not in the CTC
633 * provided by the TMA packet. Fix-up the last_mtc calculated from the TMA
634 * packet by copying the missing bits from the current MTC assuming the least
635 * difference between the two, and that the current MTC comes after last_mtc.
637 static void intel_pt_fixup_last_mtc(uint32_t mtc, int mtc_shift,
640 uint32_t first_missing_bit = 1U << (16 - mtc_shift);
641 uint32_t mask = ~(first_missing_bit - 1);
643 *last_mtc |= mtc & mask;
644 if (*last_mtc >= mtc) {
645 *last_mtc -= first_missing_bit;
650 static int intel_pt_calc_cyc_cb(struct intel_pt_pkt_info *pkt_info)
652 struct intel_pt_decoder *decoder = pkt_info->decoder;
653 struct intel_pt_calc_cyc_to_tsc_info *data = pkt_info->data;
657 uint32_t mtc, mtc_delta, ctc, fc, ctc_rem;
659 switch (pkt_info->packet.type) {
661 case INTEL_PT_TIP_PGE:
666 case INTEL_PT_MODE_EXEC:
667 case INTEL_PT_MODE_TSX:
668 case INTEL_PT_PSBEND:
672 case INTEL_PT_PTWRITE:
673 case INTEL_PT_PTWRITE_IP:
680 mtc = pkt_info->packet.payload;
681 if (decoder->mtc_shift > 8 && data->fixup_last_mtc) {
682 data->fixup_last_mtc = false;
683 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
686 if (mtc > data->last_mtc)
687 mtc_delta = mtc - data->last_mtc;
689 mtc_delta = mtc + 256 - data->last_mtc;
690 data->ctc_delta += mtc_delta << decoder->mtc_shift;
691 data->last_mtc = mtc;
693 if (decoder->tsc_ctc_mult) {
694 timestamp = data->ctc_timestamp +
695 data->ctc_delta * decoder->tsc_ctc_mult;
697 timestamp = data->ctc_timestamp +
698 multdiv(data->ctc_delta,
699 decoder->tsc_ctc_ratio_n,
700 decoder->tsc_ctc_ratio_d);
703 if (timestamp < data->timestamp)
706 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
707 data->timestamp = timestamp;
715 * For now, do not support using TSC packets - refer
716 * intel_pt_calc_cyc_to_tsc().
720 timestamp = pkt_info->packet.payload |
721 (data->timestamp & (0xffULL << 56));
722 if (data->from_mtc && timestamp < data->timestamp &&
723 data->timestamp - timestamp < decoder->tsc_slip)
725 if (timestamp < data->timestamp)
726 timestamp += (1ULL << 56);
727 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
730 data->tsc_timestamp = timestamp;
731 data->timestamp = timestamp;
740 if (!decoder->tsc_ctc_ratio_d)
743 ctc = pkt_info->packet.payload;
744 fc = pkt_info->packet.count;
745 ctc_rem = ctc & decoder->ctc_rem_mask;
747 data->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
749 data->ctc_timestamp = data->tsc_timestamp - fc;
750 if (decoder->tsc_ctc_mult) {
751 data->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
753 data->ctc_timestamp -=
754 multdiv(ctc_rem, decoder->tsc_ctc_ratio_n,
755 decoder->tsc_ctc_ratio_d);
759 data->have_tma = true;
760 data->fixup_last_mtc = true;
765 data->cycle_cnt += pkt_info->packet.payload;
769 cbr = pkt_info->packet.payload;
770 if (data->cbr && data->cbr != cbr)
773 data->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
776 case INTEL_PT_TIP_PGD:
777 case INTEL_PT_TRACESTOP:
778 case INTEL_PT_EXSTOP:
779 case INTEL_PT_EXSTOP_IP:
784 case INTEL_PT_BAD: /* Does not happen */
789 if (!data->cbr && decoder->cbr) {
790 data->cbr = decoder->cbr;
791 data->cbr_cyc_to_tsc = decoder->cbr_cyc_to_tsc;
794 if (!data->cycle_cnt)
797 cyc_to_tsc = (double)(timestamp - decoder->timestamp) / data->cycle_cnt;
799 if (data->cbr && cyc_to_tsc > data->cbr_cyc_to_tsc &&
800 cyc_to_tsc / data->cbr_cyc_to_tsc > 1.25) {
801 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle too big (c.f. CBR-based value %g), pos " x64_fmt "\n",
802 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
806 decoder->calc_cyc_to_tsc = cyc_to_tsc;
807 decoder->have_calc_cyc_to_tsc = true;
810 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. CBR-based value %g, pos " x64_fmt "\n",
811 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
813 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. unknown CBR-based value, pos " x64_fmt "\n",
814 cyc_to_tsc, pkt_info->pos);
820 static void intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder *decoder,
823 struct intel_pt_calc_cyc_to_tsc_info data = {
826 .last_mtc = decoder->last_mtc,
827 .ctc_timestamp = decoder->ctc_timestamp,
828 .ctc_delta = decoder->ctc_delta,
829 .tsc_timestamp = decoder->tsc_timestamp,
830 .timestamp = decoder->timestamp,
831 .have_tma = decoder->have_tma,
832 .fixup_last_mtc = decoder->fixup_last_mtc,
833 .from_mtc = from_mtc,
838 * For now, do not support using TSC packets for at least the reasons:
839 * 1) timing might have stopped
840 * 2) TSC packets within PSB+ can slip against CYC packets
845 intel_pt_pkt_lookahead(decoder, intel_pt_calc_cyc_cb, &data);
848 static int intel_pt_get_next_packet(struct intel_pt_decoder *decoder)
852 decoder->last_packet_type = decoder->packet.type;
855 decoder->pos += decoder->pkt_step;
856 decoder->buf += decoder->pkt_step;
857 decoder->len -= decoder->pkt_step;
860 ret = intel_pt_get_next_data(decoder);
865 ret = intel_pt_get_packet(decoder->buf, decoder->len,
867 if (ret == INTEL_PT_NEED_MORE_BYTES &&
868 decoder->len < INTEL_PT_PKT_MAX_SZ && !decoder->next_buf) {
869 ret = intel_pt_get_split_packet(decoder);
874 return intel_pt_bad_packet(decoder);
876 decoder->pkt_len = ret;
877 decoder->pkt_step = ret;
878 intel_pt_decoder_log_packet(decoder);
879 } while (decoder->packet.type == INTEL_PT_PAD);
884 static uint64_t intel_pt_next_period(struct intel_pt_decoder *decoder)
886 uint64_t timestamp, masked_timestamp;
888 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
889 masked_timestamp = timestamp & decoder->period_mask;
890 if (decoder->continuous_period) {
891 if (masked_timestamp != decoder->last_masked_timestamp)
895 masked_timestamp = timestamp & decoder->period_mask;
896 if (masked_timestamp != decoder->last_masked_timestamp) {
897 decoder->last_masked_timestamp = masked_timestamp;
898 decoder->continuous_period = true;
901 return decoder->period_ticks - (timestamp - masked_timestamp);
904 static uint64_t intel_pt_next_sample(struct intel_pt_decoder *decoder)
906 switch (decoder->period_type) {
907 case INTEL_PT_PERIOD_INSTRUCTIONS:
908 return decoder->period - decoder->period_insn_cnt;
909 case INTEL_PT_PERIOD_TICKS:
910 return intel_pt_next_period(decoder);
911 case INTEL_PT_PERIOD_NONE:
912 case INTEL_PT_PERIOD_MTC:
918 static void intel_pt_sample_insn(struct intel_pt_decoder *decoder)
920 uint64_t timestamp, masked_timestamp;
922 switch (decoder->period_type) {
923 case INTEL_PT_PERIOD_INSTRUCTIONS:
924 decoder->period_insn_cnt = 0;
926 case INTEL_PT_PERIOD_TICKS:
927 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
928 masked_timestamp = timestamp & decoder->period_mask;
929 decoder->last_masked_timestamp = masked_timestamp;
931 case INTEL_PT_PERIOD_NONE:
932 case INTEL_PT_PERIOD_MTC:
937 decoder->state.type |= INTEL_PT_INSTRUCTION;
940 static int intel_pt_walk_insn(struct intel_pt_decoder *decoder,
941 struct intel_pt_insn *intel_pt_insn, uint64_t ip)
943 uint64_t max_insn_cnt, insn_cnt = 0;
946 if (!decoder->mtc_insn)
947 decoder->mtc_insn = true;
949 max_insn_cnt = intel_pt_next_sample(decoder);
951 err = decoder->walk_insn(intel_pt_insn, &insn_cnt, &decoder->ip, ip,
952 max_insn_cnt, decoder->data);
954 decoder->tot_insn_cnt += insn_cnt;
955 decoder->timestamp_insn_cnt += insn_cnt;
956 decoder->sample_insn_cnt += insn_cnt;
957 decoder->period_insn_cnt += insn_cnt;
960 decoder->no_progress = 0;
961 decoder->pkt_state = INTEL_PT_STATE_ERR2;
962 intel_pt_log_at("ERROR: Failed to get instruction",
969 if (ip && decoder->ip == ip) {
974 if (max_insn_cnt && insn_cnt >= max_insn_cnt)
975 intel_pt_sample_insn(decoder);
977 if (intel_pt_insn->branch == INTEL_PT_BR_NO_BRANCH) {
978 decoder->state.type = INTEL_PT_INSTRUCTION;
979 decoder->state.from_ip = decoder->ip;
980 decoder->state.to_ip = 0;
981 decoder->ip += intel_pt_insn->length;
982 err = INTEL_PT_RETURN;
986 if (intel_pt_insn->op == INTEL_PT_OP_CALL) {
987 /* Zero-length calls are excluded */
988 if (intel_pt_insn->branch != INTEL_PT_BR_UNCONDITIONAL ||
989 intel_pt_insn->rel) {
990 err = intel_pt_push(&decoder->stack, decoder->ip +
991 intel_pt_insn->length);
995 } else if (intel_pt_insn->op == INTEL_PT_OP_RET) {
996 decoder->ret_addr = intel_pt_pop(&decoder->stack);
999 if (intel_pt_insn->branch == INTEL_PT_BR_UNCONDITIONAL) {
1000 int cnt = decoder->no_progress++;
1002 decoder->state.from_ip = decoder->ip;
1003 decoder->ip += intel_pt_insn->length +
1005 decoder->state.to_ip = decoder->ip;
1006 err = INTEL_PT_RETURN;
1009 * Check for being stuck in a loop. This can happen if a
1010 * decoder error results in the decoder erroneously setting the
1011 * ip to an address that is itself in an infinite loop that
1012 * consumes no packets. When that happens, there must be an
1013 * unconditional branch.
1017 decoder->stuck_ip = decoder->state.to_ip;
1018 decoder->stuck_ip_prd = 1;
1019 decoder->stuck_ip_cnt = 1;
1020 } else if (cnt > INTEL_PT_MAX_LOOPS ||
1021 decoder->state.to_ip == decoder->stuck_ip) {
1022 intel_pt_log_at("ERROR: Never-ending loop",
1023 decoder->state.to_ip);
1024 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1027 } else if (!--decoder->stuck_ip_cnt) {
1028 decoder->stuck_ip_prd += 1;
1029 decoder->stuck_ip_cnt = decoder->stuck_ip_prd;
1030 decoder->stuck_ip = decoder->state.to_ip;
1033 goto out_no_progress;
1036 decoder->no_progress = 0;
1038 decoder->state.insn_op = intel_pt_insn->op;
1039 decoder->state.insn_len = intel_pt_insn->length;
1040 memcpy(decoder->state.insn, intel_pt_insn->buf,
1041 INTEL_PT_INSN_BUF_SZ);
1043 if (decoder->tx_flags & INTEL_PT_IN_TX)
1044 decoder->state.flags |= INTEL_PT_IN_TX;
1049 static bool intel_pt_fup_event(struct intel_pt_decoder *decoder)
1053 if (decoder->set_fup_tx_flags) {
1054 decoder->set_fup_tx_flags = false;
1055 decoder->tx_flags = decoder->fup_tx_flags;
1056 decoder->state.type = INTEL_PT_TRANSACTION;
1057 decoder->state.from_ip = decoder->ip;
1058 decoder->state.to_ip = 0;
1059 decoder->state.flags = decoder->fup_tx_flags;
1062 if (decoder->set_fup_ptw) {
1063 decoder->set_fup_ptw = false;
1064 decoder->state.type = INTEL_PT_PTW;
1065 decoder->state.flags |= INTEL_PT_FUP_IP;
1066 decoder->state.from_ip = decoder->ip;
1067 decoder->state.to_ip = 0;
1068 decoder->state.ptw_payload = decoder->fup_ptw_payload;
1071 if (decoder->set_fup_mwait) {
1072 decoder->set_fup_mwait = false;
1073 decoder->state.type = INTEL_PT_MWAIT_OP;
1074 decoder->state.from_ip = decoder->ip;
1075 decoder->state.to_ip = 0;
1076 decoder->state.mwait_payload = decoder->fup_mwait_payload;
1079 if (decoder->set_fup_pwre) {
1080 decoder->set_fup_pwre = false;
1081 decoder->state.type |= INTEL_PT_PWR_ENTRY;
1082 decoder->state.type &= ~INTEL_PT_BRANCH;
1083 decoder->state.from_ip = decoder->ip;
1084 decoder->state.to_ip = 0;
1085 decoder->state.pwre_payload = decoder->fup_pwre_payload;
1088 if (decoder->set_fup_exstop) {
1089 decoder->set_fup_exstop = false;
1090 decoder->state.type |= INTEL_PT_EX_STOP;
1091 decoder->state.type &= ~INTEL_PT_BRANCH;
1092 decoder->state.flags |= INTEL_PT_FUP_IP;
1093 decoder->state.from_ip = decoder->ip;
1094 decoder->state.to_ip = 0;
1100 static int intel_pt_walk_fup(struct intel_pt_decoder *decoder)
1102 struct intel_pt_insn intel_pt_insn;
1106 ip = decoder->last_ip;
1109 err = intel_pt_walk_insn(decoder, &intel_pt_insn, ip);
1110 if (err == INTEL_PT_RETURN)
1112 if (err == -EAGAIN) {
1113 if (intel_pt_fup_event(decoder))
1117 decoder->set_fup_tx_flags = false;
1121 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1122 intel_pt_log_at("ERROR: Unexpected indirect branch",
1124 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1128 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1129 intel_pt_log_at("ERROR: Unexpected conditional branch",
1131 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1135 intel_pt_bug(decoder);
1139 static int intel_pt_walk_tip(struct intel_pt_decoder *decoder)
1141 struct intel_pt_insn intel_pt_insn;
1144 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1145 if (err == INTEL_PT_RETURN &&
1147 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1148 (decoder->state.type & INTEL_PT_BRANCH) &&
1149 decoder->pgd_ip(decoder->state.to_ip, decoder->data)) {
1150 /* Unconditional branch leaving filter region */
1151 decoder->no_progress = 0;
1152 decoder->pge = false;
1153 decoder->continuous_period = false;
1154 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1155 decoder->state.to_ip = 0;
1158 if (err == INTEL_PT_RETURN)
1163 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1164 if (decoder->pkt_state == INTEL_PT_STATE_TIP_PGD) {
1165 decoder->pge = false;
1166 decoder->continuous_period = false;
1167 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1168 decoder->state.from_ip = decoder->ip;
1169 decoder->state.to_ip = 0;
1170 if (decoder->packet.count != 0)
1171 decoder->ip = decoder->last_ip;
1173 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1174 decoder->state.from_ip = decoder->ip;
1175 if (decoder->packet.count == 0) {
1176 decoder->state.to_ip = 0;
1178 decoder->state.to_ip = decoder->last_ip;
1179 decoder->ip = decoder->last_ip;
1185 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1186 uint64_t to_ip = decoder->ip + intel_pt_insn.length +
1189 if (decoder->pgd_ip &&
1190 decoder->pkt_state == INTEL_PT_STATE_TIP_PGD &&
1191 decoder->pgd_ip(to_ip, decoder->data)) {
1192 /* Conditional branch leaving filter region */
1193 decoder->pge = false;
1194 decoder->continuous_period = false;
1195 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1196 decoder->ip = to_ip;
1197 decoder->state.from_ip = decoder->ip;
1198 decoder->state.to_ip = 0;
1201 intel_pt_log_at("ERROR: Conditional branch when expecting indirect branch",
1203 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1207 return intel_pt_bug(decoder);
1210 static int intel_pt_walk_tnt(struct intel_pt_decoder *decoder)
1212 struct intel_pt_insn intel_pt_insn;
1216 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1217 if (err == INTEL_PT_RETURN)
1222 if (intel_pt_insn.op == INTEL_PT_OP_RET) {
1223 if (!decoder->return_compression) {
1224 intel_pt_log_at("ERROR: RET when expecting conditional branch",
1226 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1229 if (!decoder->ret_addr) {
1230 intel_pt_log_at("ERROR: Bad RET compression (stack empty)",
1232 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1235 if (!(decoder->tnt.payload & BIT63)) {
1236 intel_pt_log_at("ERROR: Bad RET compression (TNT=N)",
1238 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1241 decoder->tnt.count -= 1;
1242 if (!decoder->tnt.count)
1243 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1244 decoder->tnt.payload <<= 1;
1245 decoder->state.from_ip = decoder->ip;
1246 decoder->ip = decoder->ret_addr;
1247 decoder->state.to_ip = decoder->ip;
1251 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1252 /* Handle deferred TIPs */
1253 err = intel_pt_get_next_packet(decoder);
1256 if (decoder->packet.type != INTEL_PT_TIP ||
1257 decoder->packet.count == 0) {
1258 intel_pt_log_at("ERROR: Missing deferred TIP for indirect branch",
1260 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1261 decoder->pkt_step = 0;
1264 intel_pt_set_last_ip(decoder);
1265 decoder->state.from_ip = decoder->ip;
1266 decoder->state.to_ip = decoder->last_ip;
1267 decoder->ip = decoder->last_ip;
1271 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1272 decoder->tnt.count -= 1;
1273 if (!decoder->tnt.count)
1274 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1275 if (decoder->tnt.payload & BIT63) {
1276 decoder->tnt.payload <<= 1;
1277 decoder->state.from_ip = decoder->ip;
1278 decoder->ip += intel_pt_insn.length +
1280 decoder->state.to_ip = decoder->ip;
1283 /* Instruction sample for a non-taken branch */
1284 if (decoder->state.type & INTEL_PT_INSTRUCTION) {
1285 decoder->tnt.payload <<= 1;
1286 decoder->state.type = INTEL_PT_INSTRUCTION;
1287 decoder->state.from_ip = decoder->ip;
1288 decoder->state.to_ip = 0;
1289 decoder->ip += intel_pt_insn.length;
1292 decoder->ip += intel_pt_insn.length;
1293 if (!decoder->tnt.count)
1295 decoder->tnt.payload <<= 1;
1299 return intel_pt_bug(decoder);
1303 static int intel_pt_mode_tsx(struct intel_pt_decoder *decoder, bool *no_tip)
1305 unsigned int fup_tx_flags;
1308 fup_tx_flags = decoder->packet.payload &
1309 (INTEL_PT_IN_TX | INTEL_PT_ABORT_TX);
1310 err = intel_pt_get_next_packet(decoder);
1313 if (decoder->packet.type == INTEL_PT_FUP) {
1314 decoder->fup_tx_flags = fup_tx_flags;
1315 decoder->set_fup_tx_flags = true;
1316 if (!(decoder->fup_tx_flags & INTEL_PT_ABORT_TX))
1319 intel_pt_log_at("ERROR: Missing FUP after MODE.TSX",
1321 intel_pt_update_in_tx(decoder);
1326 static void intel_pt_calc_tsc_timestamp(struct intel_pt_decoder *decoder)
1330 decoder->have_tma = false;
1332 if (decoder->ref_timestamp) {
1333 timestamp = decoder->packet.payload |
1334 (decoder->ref_timestamp & (0xffULL << 56));
1335 if (timestamp < decoder->ref_timestamp) {
1336 if (decoder->ref_timestamp - timestamp > (1ULL << 55))
1337 timestamp += (1ULL << 56);
1339 if (timestamp - decoder->ref_timestamp > (1ULL << 55))
1340 timestamp -= (1ULL << 56);
1342 decoder->tsc_timestamp = timestamp;
1343 decoder->timestamp = timestamp;
1344 decoder->ref_timestamp = 0;
1345 decoder->timestamp_insn_cnt = 0;
1346 } else if (decoder->timestamp) {
1347 timestamp = decoder->packet.payload |
1348 (decoder->timestamp & (0xffULL << 56));
1349 decoder->tsc_timestamp = timestamp;
1350 if (timestamp < decoder->timestamp &&
1351 decoder->timestamp - timestamp < decoder->tsc_slip) {
1352 intel_pt_log_to("Suppressing backwards timestamp",
1354 timestamp = decoder->timestamp;
1356 if (timestamp < decoder->timestamp) {
1357 intel_pt_log_to("Wraparound timestamp", timestamp);
1358 timestamp += (1ULL << 56);
1359 decoder->tsc_timestamp = timestamp;
1361 decoder->timestamp = timestamp;
1362 decoder->timestamp_insn_cnt = 0;
1365 if (decoder->last_packet_type == INTEL_PT_CYC) {
1366 decoder->cyc_ref_timestamp = decoder->timestamp;
1367 decoder->cycle_cnt = 0;
1368 decoder->have_calc_cyc_to_tsc = false;
1369 intel_pt_calc_cyc_to_tsc(decoder, false);
1372 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1375 static int intel_pt_overflow(struct intel_pt_decoder *decoder)
1377 intel_pt_log("ERROR: Buffer overflow\n");
1378 intel_pt_clear_tx_flags(decoder);
1379 decoder->have_tma = false;
1381 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1382 decoder->overflow = true;
1386 static void intel_pt_calc_tma(struct intel_pt_decoder *decoder)
1388 uint32_t ctc = decoder->packet.payload;
1389 uint32_t fc = decoder->packet.count;
1390 uint32_t ctc_rem = ctc & decoder->ctc_rem_mask;
1392 if (!decoder->tsc_ctc_ratio_d)
1395 decoder->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
1396 decoder->ctc_timestamp = decoder->tsc_timestamp - fc;
1397 if (decoder->tsc_ctc_mult) {
1398 decoder->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
1400 decoder->ctc_timestamp -= multdiv(ctc_rem,
1401 decoder->tsc_ctc_ratio_n,
1402 decoder->tsc_ctc_ratio_d);
1404 decoder->ctc_delta = 0;
1405 decoder->have_tma = true;
1406 decoder->fixup_last_mtc = true;
1407 intel_pt_log("CTC timestamp " x64_fmt " last MTC %#x CTC rem %#x\n",
1408 decoder->ctc_timestamp, decoder->last_mtc, ctc_rem);
1411 static void intel_pt_calc_mtc_timestamp(struct intel_pt_decoder *decoder)
1414 uint32_t mtc, mtc_delta;
1416 if (!decoder->have_tma)
1419 mtc = decoder->packet.payload;
1421 if (decoder->mtc_shift > 8 && decoder->fixup_last_mtc) {
1422 decoder->fixup_last_mtc = false;
1423 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
1424 &decoder->last_mtc);
1427 if (mtc > decoder->last_mtc)
1428 mtc_delta = mtc - decoder->last_mtc;
1430 mtc_delta = mtc + 256 - decoder->last_mtc;
1432 decoder->ctc_delta += mtc_delta << decoder->mtc_shift;
1434 if (decoder->tsc_ctc_mult) {
1435 timestamp = decoder->ctc_timestamp +
1436 decoder->ctc_delta * decoder->tsc_ctc_mult;
1438 timestamp = decoder->ctc_timestamp +
1439 multdiv(decoder->ctc_delta,
1440 decoder->tsc_ctc_ratio_n,
1441 decoder->tsc_ctc_ratio_d);
1444 if (timestamp < decoder->timestamp)
1445 intel_pt_log("Suppressing MTC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1446 timestamp, decoder->timestamp);
1448 decoder->timestamp = timestamp;
1450 decoder->timestamp_insn_cnt = 0;
1451 decoder->last_mtc = mtc;
1453 if (decoder->last_packet_type == INTEL_PT_CYC) {
1454 decoder->cyc_ref_timestamp = decoder->timestamp;
1455 decoder->cycle_cnt = 0;
1456 decoder->have_calc_cyc_to_tsc = false;
1457 intel_pt_calc_cyc_to_tsc(decoder, true);
1461 static void intel_pt_calc_cbr(struct intel_pt_decoder *decoder)
1463 unsigned int cbr = decoder->packet.payload & 0xff;
1465 decoder->cbr_payload = decoder->packet.payload;
1467 if (decoder->cbr == cbr)
1471 decoder->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
1474 static void intel_pt_calc_cyc_timestamp(struct intel_pt_decoder *decoder)
1476 uint64_t timestamp = decoder->cyc_ref_timestamp;
1478 decoder->have_cyc = true;
1480 decoder->cycle_cnt += decoder->packet.payload;
1482 if (!decoder->cyc_ref_timestamp)
1485 if (decoder->have_calc_cyc_to_tsc)
1486 timestamp += decoder->cycle_cnt * decoder->calc_cyc_to_tsc;
1487 else if (decoder->cbr)
1488 timestamp += decoder->cycle_cnt * decoder->cbr_cyc_to_tsc;
1492 if (timestamp < decoder->timestamp)
1493 intel_pt_log("Suppressing CYC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1494 timestamp, decoder->timestamp);
1496 decoder->timestamp = timestamp;
1498 decoder->timestamp_insn_cnt = 0;
1501 /* Walk PSB+ packets when already in sync. */
1502 static int intel_pt_walk_psbend(struct intel_pt_decoder *decoder)
1507 err = intel_pt_get_next_packet(decoder);
1511 switch (decoder->packet.type) {
1512 case INTEL_PT_PSBEND:
1515 case INTEL_PT_TIP_PGD:
1516 case INTEL_PT_TIP_PGE:
1519 case INTEL_PT_TRACESTOP:
1522 case INTEL_PT_PTWRITE:
1523 case INTEL_PT_PTWRITE_IP:
1524 case INTEL_PT_EXSTOP:
1525 case INTEL_PT_EXSTOP_IP:
1526 case INTEL_PT_MWAIT:
1529 decoder->have_tma = false;
1530 intel_pt_log("ERROR: Unexpected packet\n");
1534 return intel_pt_overflow(decoder);
1537 intel_pt_calc_tsc_timestamp(decoder);
1541 intel_pt_calc_tma(decoder);
1545 intel_pt_calc_cbr(decoder);
1548 case INTEL_PT_MODE_EXEC:
1549 decoder->exec_mode = decoder->packet.payload;
1553 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1557 decoder->pge = true;
1558 if (decoder->packet.count)
1559 intel_pt_set_last_ip(decoder);
1562 case INTEL_PT_MODE_TSX:
1563 intel_pt_update_in_tx(decoder);
1567 intel_pt_calc_mtc_timestamp(decoder);
1568 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1569 decoder->state.type |= INTEL_PT_INSTRUCTION;
1582 static int intel_pt_walk_fup_tip(struct intel_pt_decoder *decoder)
1586 if (decoder->tx_flags & INTEL_PT_ABORT_TX) {
1587 decoder->tx_flags = 0;
1588 decoder->state.flags &= ~INTEL_PT_IN_TX;
1589 decoder->state.flags |= INTEL_PT_ABORT_TX;
1591 decoder->state.flags |= INTEL_PT_ASYNC;
1595 err = intel_pt_get_next_packet(decoder);
1599 switch (decoder->packet.type) {
1602 case INTEL_PT_TRACESTOP:
1607 case INTEL_PT_MODE_TSX:
1609 case INTEL_PT_PSBEND:
1610 case INTEL_PT_PTWRITE:
1611 case INTEL_PT_PTWRITE_IP:
1612 case INTEL_PT_EXSTOP:
1613 case INTEL_PT_EXSTOP_IP:
1614 case INTEL_PT_MWAIT:
1617 intel_pt_log("ERROR: Missing TIP after FUP\n");
1618 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1622 return intel_pt_overflow(decoder);
1624 case INTEL_PT_TIP_PGD:
1625 decoder->state.from_ip = decoder->ip;
1626 decoder->state.to_ip = 0;
1627 if (decoder->packet.count != 0) {
1628 intel_pt_set_ip(decoder);
1629 intel_pt_log("Omitting PGD ip " x64_fmt "\n",
1632 decoder->pge = false;
1633 decoder->continuous_period = false;
1636 case INTEL_PT_TIP_PGE:
1637 decoder->pge = true;
1638 intel_pt_log("Omitting PGE ip " x64_fmt "\n",
1640 decoder->state.from_ip = 0;
1641 if (decoder->packet.count == 0) {
1642 decoder->state.to_ip = 0;
1644 intel_pt_set_ip(decoder);
1645 decoder->state.to_ip = decoder->ip;
1650 decoder->state.from_ip = decoder->ip;
1651 if (decoder->packet.count == 0) {
1652 decoder->state.to_ip = 0;
1654 intel_pt_set_ip(decoder);
1655 decoder->state.to_ip = decoder->ip;
1660 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1664 intel_pt_calc_mtc_timestamp(decoder);
1665 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1666 decoder->state.type |= INTEL_PT_INSTRUCTION;
1670 intel_pt_calc_cyc_timestamp(decoder);
1673 case INTEL_PT_MODE_EXEC:
1674 decoder->exec_mode = decoder->packet.payload;
1683 return intel_pt_bug(decoder);
1688 static int intel_pt_walk_trace(struct intel_pt_decoder *decoder)
1690 bool no_tip = false;
1694 err = intel_pt_get_next_packet(decoder);
1698 switch (decoder->packet.type) {
1700 if (!decoder->packet.count)
1702 decoder->tnt = decoder->packet;
1703 decoder->pkt_state = INTEL_PT_STATE_TNT;
1704 err = intel_pt_walk_tnt(decoder);
1709 case INTEL_PT_TIP_PGD:
1710 if (decoder->packet.count != 0)
1711 intel_pt_set_last_ip(decoder);
1712 decoder->pkt_state = INTEL_PT_STATE_TIP_PGD;
1713 return intel_pt_walk_tip(decoder);
1715 case INTEL_PT_TIP_PGE: {
1716 decoder->pge = true;
1717 if (decoder->packet.count == 0) {
1718 intel_pt_log_at("Skipping zero TIP.PGE",
1722 intel_pt_set_ip(decoder);
1723 decoder->state.from_ip = 0;
1724 decoder->state.to_ip = decoder->ip;
1729 return intel_pt_overflow(decoder);
1732 if (decoder->packet.count != 0)
1733 intel_pt_set_last_ip(decoder);
1734 decoder->pkt_state = INTEL_PT_STATE_TIP;
1735 return intel_pt_walk_tip(decoder);
1738 if (decoder->packet.count == 0) {
1739 intel_pt_log_at("Skipping zero FUP",
1744 intel_pt_set_last_ip(decoder);
1745 if (!decoder->branch_enable) {
1746 decoder->ip = decoder->last_ip;
1747 if (intel_pt_fup_event(decoder))
1752 if (decoder->set_fup_mwait)
1754 err = intel_pt_walk_fup(decoder);
1755 if (err != -EAGAIN) {
1759 decoder->pkt_state =
1760 INTEL_PT_STATE_FUP_NO_TIP;
1762 decoder->pkt_state = INTEL_PT_STATE_FUP;
1769 return intel_pt_walk_fup_tip(decoder);
1771 case INTEL_PT_TRACESTOP:
1772 decoder->pge = false;
1773 decoder->continuous_period = false;
1774 intel_pt_clear_tx_flags(decoder);
1775 decoder->have_tma = false;
1779 decoder->last_ip = 0;
1780 decoder->have_last_ip = true;
1781 intel_pt_clear_stack(&decoder->stack);
1782 err = intel_pt_walk_psbend(decoder);
1790 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1794 intel_pt_calc_mtc_timestamp(decoder);
1795 if (decoder->period_type != INTEL_PT_PERIOD_MTC)
1798 * Ensure that there has been an instruction since the
1801 if (!decoder->mtc_insn)
1803 decoder->mtc_insn = false;
1804 /* Ensure that there is a timestamp */
1805 if (!decoder->timestamp)
1807 decoder->state.type = INTEL_PT_INSTRUCTION;
1808 decoder->state.from_ip = decoder->ip;
1809 decoder->state.to_ip = 0;
1810 decoder->mtc_insn = false;
1814 intel_pt_calc_tsc_timestamp(decoder);
1818 intel_pt_calc_tma(decoder);
1822 intel_pt_calc_cyc_timestamp(decoder);
1826 intel_pt_calc_cbr(decoder);
1827 if (!decoder->branch_enable &&
1828 decoder->cbr != decoder->cbr_seen) {
1829 decoder->cbr_seen = decoder->cbr;
1830 decoder->state.type = INTEL_PT_CBR_CHG;
1831 decoder->state.from_ip = decoder->ip;
1832 decoder->state.to_ip = 0;
1833 decoder->state.cbr_payload =
1834 decoder->packet.payload;
1839 case INTEL_PT_MODE_EXEC:
1840 decoder->exec_mode = decoder->packet.payload;
1843 case INTEL_PT_MODE_TSX:
1844 /* MODE_TSX need not be followed by FUP */
1845 if (!decoder->pge) {
1846 intel_pt_update_in_tx(decoder);
1849 err = intel_pt_mode_tsx(decoder, &no_tip);
1854 case INTEL_PT_BAD: /* Does not happen */
1855 return intel_pt_bug(decoder);
1857 case INTEL_PT_PSBEND:
1863 case INTEL_PT_PTWRITE_IP:
1864 decoder->fup_ptw_payload = decoder->packet.payload;
1865 err = intel_pt_get_next_packet(decoder);
1868 if (decoder->packet.type == INTEL_PT_FUP) {
1869 decoder->set_fup_ptw = true;
1872 intel_pt_log_at("ERROR: Missing FUP after PTWRITE",
1877 case INTEL_PT_PTWRITE:
1878 decoder->state.type = INTEL_PT_PTW;
1879 decoder->state.from_ip = decoder->ip;
1880 decoder->state.to_ip = 0;
1881 decoder->state.ptw_payload = decoder->packet.payload;
1884 case INTEL_PT_MWAIT:
1885 decoder->fup_mwait_payload = decoder->packet.payload;
1886 decoder->set_fup_mwait = true;
1890 if (decoder->set_fup_mwait) {
1891 decoder->fup_pwre_payload =
1892 decoder->packet.payload;
1893 decoder->set_fup_pwre = true;
1896 decoder->state.type = INTEL_PT_PWR_ENTRY;
1897 decoder->state.from_ip = decoder->ip;
1898 decoder->state.to_ip = 0;
1899 decoder->state.pwrx_payload = decoder->packet.payload;
1902 case INTEL_PT_EXSTOP_IP:
1903 err = intel_pt_get_next_packet(decoder);
1906 if (decoder->packet.type == INTEL_PT_FUP) {
1907 decoder->set_fup_exstop = true;
1910 intel_pt_log_at("ERROR: Missing FUP after EXSTOP",
1915 case INTEL_PT_EXSTOP:
1916 decoder->state.type = INTEL_PT_EX_STOP;
1917 decoder->state.from_ip = decoder->ip;
1918 decoder->state.to_ip = 0;
1922 decoder->state.type = INTEL_PT_PWR_EXIT;
1923 decoder->state.from_ip = decoder->ip;
1924 decoder->state.to_ip = 0;
1925 decoder->state.pwrx_payload = decoder->packet.payload;
1929 return intel_pt_bug(decoder);
1934 static inline bool intel_pt_have_ip(struct intel_pt_decoder *decoder)
1936 return decoder->packet.count &&
1937 (decoder->have_last_ip || decoder->packet.count == 3 ||
1938 decoder->packet.count == 6);
1941 /* Walk PSB+ packets to get in sync. */
1942 static int intel_pt_walk_psb(struct intel_pt_decoder *decoder)
1947 err = intel_pt_get_next_packet(decoder);
1951 switch (decoder->packet.type) {
1952 case INTEL_PT_TIP_PGD:
1953 decoder->continuous_period = false;
1955 case INTEL_PT_TIP_PGE:
1957 case INTEL_PT_PTWRITE:
1958 case INTEL_PT_PTWRITE_IP:
1959 case INTEL_PT_EXSTOP:
1960 case INTEL_PT_EXSTOP_IP:
1961 case INTEL_PT_MWAIT:
1964 intel_pt_log("ERROR: Unexpected packet\n");
1968 decoder->pge = true;
1969 if (intel_pt_have_ip(decoder)) {
1970 uint64_t current_ip = decoder->ip;
1972 intel_pt_set_ip(decoder);
1974 intel_pt_log_to("Setting IP",
1980 intel_pt_calc_mtc_timestamp(decoder);
1984 intel_pt_calc_tsc_timestamp(decoder);
1988 intel_pt_calc_tma(decoder);
1992 intel_pt_calc_cyc_timestamp(decoder);
1996 intel_pt_calc_cbr(decoder);
2000 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
2003 case INTEL_PT_MODE_EXEC:
2004 decoder->exec_mode = decoder->packet.payload;
2007 case INTEL_PT_MODE_TSX:
2008 intel_pt_update_in_tx(decoder);
2011 case INTEL_PT_TRACESTOP:
2012 decoder->pge = false;
2013 decoder->continuous_period = false;
2014 intel_pt_clear_tx_flags(decoder);
2018 decoder->have_tma = false;
2019 intel_pt_log("ERROR: Unexpected packet\n");
2021 decoder->pkt_state = INTEL_PT_STATE_ERR4;
2023 decoder->pkt_state = INTEL_PT_STATE_ERR3;
2026 case INTEL_PT_BAD: /* Does not happen */
2027 return intel_pt_bug(decoder);
2030 return intel_pt_overflow(decoder);
2032 case INTEL_PT_PSBEND:
2045 static int intel_pt_walk_to_ip(struct intel_pt_decoder *decoder)
2050 err = intel_pt_get_next_packet(decoder);
2054 switch (decoder->packet.type) {
2055 case INTEL_PT_TIP_PGD:
2056 decoder->continuous_period = false;
2058 case INTEL_PT_TIP_PGE:
2060 decoder->pge = decoder->packet.type != INTEL_PT_TIP_PGD;
2061 if (intel_pt_have_ip(decoder))
2062 intel_pt_set_ip(decoder);
2068 if (intel_pt_have_ip(decoder))
2069 intel_pt_set_ip(decoder);
2075 intel_pt_calc_mtc_timestamp(decoder);
2079 intel_pt_calc_tsc_timestamp(decoder);
2083 intel_pt_calc_tma(decoder);
2087 intel_pt_calc_cyc_timestamp(decoder);
2091 intel_pt_calc_cbr(decoder);
2095 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
2098 case INTEL_PT_MODE_EXEC:
2099 decoder->exec_mode = decoder->packet.payload;
2102 case INTEL_PT_MODE_TSX:
2103 intel_pt_update_in_tx(decoder);
2107 return intel_pt_overflow(decoder);
2109 case INTEL_PT_BAD: /* Does not happen */
2110 return intel_pt_bug(decoder);
2112 case INTEL_PT_TRACESTOP:
2113 decoder->pge = false;
2114 decoder->continuous_period = false;
2115 intel_pt_clear_tx_flags(decoder);
2116 decoder->have_tma = false;
2120 decoder->last_ip = 0;
2121 decoder->have_last_ip = true;
2122 intel_pt_clear_stack(&decoder->stack);
2123 err = intel_pt_walk_psb(decoder);
2127 /* Do not have a sample */
2128 decoder->state.type = 0;
2134 case INTEL_PT_PSBEND:
2138 case INTEL_PT_PTWRITE:
2139 case INTEL_PT_PTWRITE_IP:
2140 case INTEL_PT_EXSTOP:
2141 case INTEL_PT_EXSTOP_IP:
2142 case INTEL_PT_MWAIT:
2151 static int intel_pt_sync_ip(struct intel_pt_decoder *decoder)
2155 decoder->set_fup_tx_flags = false;
2156 decoder->set_fup_ptw = false;
2157 decoder->set_fup_mwait = false;
2158 decoder->set_fup_pwre = false;
2159 decoder->set_fup_exstop = false;
2161 if (!decoder->branch_enable) {
2162 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2163 decoder->overflow = false;
2164 decoder->state.type = 0; /* Do not have a sample */
2168 intel_pt_log("Scanning for full IP\n");
2169 err = intel_pt_walk_to_ip(decoder);
2173 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2174 decoder->overflow = false;
2176 decoder->state.from_ip = 0;
2177 decoder->state.to_ip = decoder->ip;
2178 intel_pt_log_to("Setting IP", decoder->ip);
2183 static int intel_pt_part_psb(struct intel_pt_decoder *decoder)
2185 const unsigned char *end = decoder->buf + decoder->len;
2188 for (i = INTEL_PT_PSB_LEN - 1; i; i--) {
2189 if (i > decoder->len)
2191 if (!memcmp(end - i, INTEL_PT_PSB_STR, i))
2197 static int intel_pt_rest_psb(struct intel_pt_decoder *decoder, int part_psb)
2199 size_t rest_psb = INTEL_PT_PSB_LEN - part_psb;
2200 const char *psb = INTEL_PT_PSB_STR;
2202 if (rest_psb > decoder->len ||
2203 memcmp(decoder->buf, psb + part_psb, rest_psb))
2209 static int intel_pt_get_split_psb(struct intel_pt_decoder *decoder,
2214 decoder->pos += decoder->len;
2217 ret = intel_pt_get_next_data(decoder);
2221 rest_psb = intel_pt_rest_psb(decoder, part_psb);
2225 decoder->pos -= part_psb;
2226 decoder->next_buf = decoder->buf + rest_psb;
2227 decoder->next_len = decoder->len - rest_psb;
2228 memcpy(decoder->temp_buf, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2229 decoder->buf = decoder->temp_buf;
2230 decoder->len = INTEL_PT_PSB_LEN;
2235 static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder)
2237 unsigned char *next;
2240 intel_pt_log("Scanning for PSB\n");
2242 if (!decoder->len) {
2243 ret = intel_pt_get_next_data(decoder);
2248 next = memmem(decoder->buf, decoder->len, INTEL_PT_PSB_STR,
2253 part_psb = intel_pt_part_psb(decoder);
2255 ret = intel_pt_get_split_psb(decoder, part_psb);
2259 decoder->pos += decoder->len;
2265 decoder->pkt_step = next - decoder->buf;
2266 return intel_pt_get_next_packet(decoder);
2270 static int intel_pt_sync(struct intel_pt_decoder *decoder)
2274 decoder->pge = false;
2275 decoder->continuous_period = false;
2276 decoder->have_last_ip = false;
2277 decoder->last_ip = 0;
2279 intel_pt_clear_stack(&decoder->stack);
2281 err = intel_pt_scan_for_psb(decoder);
2285 decoder->have_last_ip = true;
2286 decoder->pkt_state = INTEL_PT_STATE_NO_IP;
2288 err = intel_pt_walk_psb(decoder);
2293 decoder->state.type = 0; /* Do not have a sample */
2294 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2296 return intel_pt_sync_ip(decoder);
2302 static uint64_t intel_pt_est_timestamp(struct intel_pt_decoder *decoder)
2304 uint64_t est = decoder->sample_insn_cnt << 1;
2306 if (!decoder->cbr || !decoder->max_non_turbo_ratio)
2309 est *= decoder->max_non_turbo_ratio;
2310 est /= decoder->cbr;
2312 return decoder->sample_timestamp + est;
2315 const struct intel_pt_state *intel_pt_decode(struct intel_pt_decoder *decoder)
2320 decoder->state.type = INTEL_PT_BRANCH;
2321 decoder->state.flags = 0;
2323 switch (decoder->pkt_state) {
2324 case INTEL_PT_STATE_NO_PSB:
2325 err = intel_pt_sync(decoder);
2327 case INTEL_PT_STATE_NO_IP:
2328 decoder->have_last_ip = false;
2329 decoder->last_ip = 0;
2332 case INTEL_PT_STATE_ERR_RESYNC:
2333 err = intel_pt_sync_ip(decoder);
2335 case INTEL_PT_STATE_IN_SYNC:
2336 err = intel_pt_walk_trace(decoder);
2338 case INTEL_PT_STATE_TNT:
2339 err = intel_pt_walk_tnt(decoder);
2341 err = intel_pt_walk_trace(decoder);
2343 case INTEL_PT_STATE_TIP:
2344 case INTEL_PT_STATE_TIP_PGD:
2345 err = intel_pt_walk_tip(decoder);
2347 case INTEL_PT_STATE_FUP:
2348 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2349 err = intel_pt_walk_fup(decoder);
2351 err = intel_pt_walk_fup_tip(decoder);
2353 decoder->pkt_state = INTEL_PT_STATE_FUP;
2355 case INTEL_PT_STATE_FUP_NO_TIP:
2356 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2357 err = intel_pt_walk_fup(decoder);
2359 err = intel_pt_walk_trace(decoder);
2362 err = intel_pt_bug(decoder);
2365 } while (err == -ENOLINK);
2368 decoder->state.err = intel_pt_ext_err(err);
2369 decoder->state.from_ip = decoder->ip;
2370 decoder->sample_timestamp = decoder->timestamp;
2371 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
2373 decoder->state.err = 0;
2374 if (decoder->cbr != decoder->cbr_seen && decoder->state.type) {
2375 decoder->cbr_seen = decoder->cbr;
2376 decoder->state.type |= INTEL_PT_CBR_CHG;
2377 decoder->state.cbr_payload = decoder->cbr_payload;
2379 if (intel_pt_sample_time(decoder->pkt_state)) {
2380 decoder->sample_timestamp = decoder->timestamp;
2381 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
2385 decoder->state.timestamp = decoder->sample_timestamp;
2386 decoder->state.est_timestamp = intel_pt_est_timestamp(decoder);
2387 decoder->state.cr3 = decoder->cr3;
2388 decoder->state.tot_insn_cnt = decoder->tot_insn_cnt;
2390 return &decoder->state;
2393 static bool intel_pt_at_psb(unsigned char *buf, size_t len)
2395 if (len < INTEL_PT_PSB_LEN)
2397 return memmem(buf, INTEL_PT_PSB_LEN, INTEL_PT_PSB_STR,
2402 * intel_pt_next_psb - move buffer pointer to the start of the next PSB packet.
2403 * @buf: pointer to buffer pointer
2404 * @len: size of buffer
2406 * Updates the buffer pointer to point to the start of the next PSB packet if
2407 * there is one, otherwise the buffer pointer is unchanged. If @buf is updated,
2408 * @len is adjusted accordingly.
2410 * Return: %true if a PSB packet is found, %false otherwise.
2412 static bool intel_pt_next_psb(unsigned char **buf, size_t *len)
2414 unsigned char *next;
2416 next = memmem(*buf, *len, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2418 *len -= next - *buf;
2426 * intel_pt_step_psb - move buffer pointer to the start of the following PSB
2428 * @buf: pointer to buffer pointer
2429 * @len: size of buffer
2431 * Updates the buffer pointer to point to the start of the following PSB packet
2432 * (skipping the PSB at @buf itself) if there is one, otherwise the buffer
2433 * pointer is unchanged. If @buf is updated, @len is adjusted accordingly.
2435 * Return: %true if a PSB packet is found, %false otherwise.
2437 static bool intel_pt_step_psb(unsigned char **buf, size_t *len)
2439 unsigned char *next;
2444 next = memmem(*buf + 1, *len - 1, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2446 *len -= next - *buf;
2454 * intel_pt_last_psb - find the last PSB packet in a buffer.
2456 * @len: size of buffer
2458 * This function finds the last PSB in a buffer.
2460 * Return: A pointer to the last PSB in @buf if found, %NULL otherwise.
2462 static unsigned char *intel_pt_last_psb(unsigned char *buf, size_t len)
2464 const char *n = INTEL_PT_PSB_STR;
2468 if (len < INTEL_PT_PSB_LEN)
2471 k = len - INTEL_PT_PSB_LEN + 1;
2473 p = memrchr(buf, n[0], k);
2476 if (!memcmp(p + 1, n + 1, INTEL_PT_PSB_LEN - 1))
2485 * intel_pt_next_tsc - find and return next TSC.
2487 * @len: size of buffer
2488 * @tsc: TSC value returned
2490 * Find a TSC packet in @buf and return the TSC value. This function assumes
2491 * that @buf starts at a PSB and that PSB+ will contain TSC and so stops if a
2492 * PSBEND packet is found.
2494 * Return: %true if TSC is found, false otherwise.
2496 static bool intel_pt_next_tsc(unsigned char *buf, size_t len, uint64_t *tsc)
2498 struct intel_pt_pkt packet;
2502 ret = intel_pt_get_packet(buf, len, &packet);
2505 if (packet.type == INTEL_PT_TSC) {
2506 *tsc = packet.payload;
2509 if (packet.type == INTEL_PT_PSBEND)
2518 * intel_pt_tsc_cmp - compare 7-byte TSCs.
2519 * @tsc1: first TSC to compare
2520 * @tsc2: second TSC to compare
2522 * This function compares 7-byte TSC values allowing for the possibility that
2523 * TSC wrapped around. Generally it is not possible to know if TSC has wrapped
2524 * around so for that purpose this function assumes the absolute difference is
2525 * less than half the maximum difference.
2527 * Return: %-1 if @tsc1 is before @tsc2, %0 if @tsc1 == @tsc2, %1 if @tsc1 is
2530 static int intel_pt_tsc_cmp(uint64_t tsc1, uint64_t tsc2)
2532 const uint64_t halfway = (1ULL << 55);
2538 if (tsc2 - tsc1 < halfway)
2543 if (tsc1 - tsc2 < halfway)
2551 * intel_pt_find_overlap_tsc - determine start of non-overlapped trace data
2553 * @buf_a: first buffer
2554 * @len_a: size of first buffer
2555 * @buf_b: second buffer
2556 * @len_b: size of second buffer
2558 * If the trace contains TSC we can look at the last TSC of @buf_a and the
2559 * first TSC of @buf_b in order to determine if the buffers overlap, and then
2560 * walk forward in @buf_b until a later TSC is found. A precondition is that
2561 * @buf_a and @buf_b are positioned at a PSB.
2563 * Return: A pointer into @buf_b from where non-overlapped data starts, or
2564 * @buf_b + @len_b if there is no non-overlapped data.
2566 static unsigned char *intel_pt_find_overlap_tsc(unsigned char *buf_a,
2568 unsigned char *buf_b,
2571 uint64_t tsc_a, tsc_b;
2575 p = intel_pt_last_psb(buf_a, len_a);
2577 return buf_b; /* No PSB in buf_a => no overlap */
2579 len = len_a - (p - buf_a);
2580 if (!intel_pt_next_tsc(p, len, &tsc_a)) {
2581 /* The last PSB+ in buf_a is incomplete, so go back one more */
2583 p = intel_pt_last_psb(buf_a, len_a);
2585 return buf_b; /* No full PSB+ => assume no overlap */
2586 len = len_a - (p - buf_a);
2587 if (!intel_pt_next_tsc(p, len, &tsc_a))
2588 return buf_b; /* No TSC in buf_a => assume no overlap */
2592 /* Ignore PSB+ with no TSC */
2593 if (intel_pt_next_tsc(buf_b, len_b, &tsc_b) &&
2594 intel_pt_tsc_cmp(tsc_a, tsc_b) < 0)
2595 return buf_b; /* tsc_a < tsc_b => no overlap */
2597 if (!intel_pt_step_psb(&buf_b, &len_b))
2598 return buf_b + len_b; /* No PSB in buf_b => no data */
2603 * intel_pt_find_overlap - determine start of non-overlapped trace data.
2604 * @buf_a: first buffer
2605 * @len_a: size of first buffer
2606 * @buf_b: second buffer
2607 * @len_b: size of second buffer
2608 * @have_tsc: can use TSC packets to detect overlap
2610 * When trace samples or snapshots are recorded there is the possibility that
2611 * the data overlaps. Note that, for the purposes of decoding, data is only
2612 * useful if it begins with a PSB packet.
2614 * Return: A pointer into @buf_b from where non-overlapped data starts, or
2615 * @buf_b + @len_b if there is no non-overlapped data.
2617 unsigned char *intel_pt_find_overlap(unsigned char *buf_a, size_t len_a,
2618 unsigned char *buf_b, size_t len_b,
2621 unsigned char *found;
2623 /* Buffer 'b' must start at PSB so throw away everything before that */
2624 if (!intel_pt_next_psb(&buf_b, &len_b))
2625 return buf_b + len_b; /* No PSB */
2627 if (!intel_pt_next_psb(&buf_a, &len_a))
2628 return buf_b; /* No overlap */
2631 found = intel_pt_find_overlap_tsc(buf_a, len_a, buf_b, len_b);
2637 * Buffer 'b' cannot end within buffer 'a' so, for comparison purposes,
2638 * we can ignore the first part of buffer 'a'.
2640 while (len_b < len_a) {
2641 if (!intel_pt_step_psb(&buf_a, &len_a))
2642 return buf_b; /* No overlap */
2645 /* Now len_b >= len_a */
2646 if (len_b > len_a) {
2647 /* The leftover buffer 'b' must start at a PSB */
2648 while (!intel_pt_at_psb(buf_b + len_a, len_b - len_a)) {
2649 if (!intel_pt_step_psb(&buf_a, &len_a))
2650 return buf_b; /* No overlap */
2655 /* Potential overlap so check the bytes */
2656 found = memmem(buf_a, len_a, buf_b, len_a);
2658 return buf_b + len_a;
2660 /* Try again at next PSB in buffer 'a' */
2661 if (!intel_pt_step_psb(&buf_a, &len_a))
2662 return buf_b; /* No overlap */
2664 /* The leftover buffer 'b' must start at a PSB */
2665 while (!intel_pt_at_psb(buf_b + len_a, len_b - len_a)) {
2666 if (!intel_pt_step_psb(&buf_a, &len_a))
2667 return buf_b; /* No overlap */