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>
29 #include "../auxtrace.h"
31 #include "intel-pt-insn-decoder.h"
32 #include "intel-pt-pkt-decoder.h"
33 #include "intel-pt-decoder.h"
34 #include "intel-pt-log.h"
36 #define INTEL_PT_BLK_SIZE 1024
38 #define BIT63 (((uint64_t)1 << 63))
40 #define INTEL_PT_RETURN 1
42 /* Maximum number of loops with no packets consumed i.e. stuck in a loop */
43 #define INTEL_PT_MAX_LOOPS 10000
46 struct intel_pt_blk *prev;
47 uint64_t ip[INTEL_PT_BLK_SIZE];
50 struct intel_pt_stack {
51 struct intel_pt_blk *blk;
52 struct intel_pt_blk *spare;
56 enum intel_pt_pkt_state {
57 INTEL_PT_STATE_NO_PSB,
59 INTEL_PT_STATE_ERR_RESYNC,
60 INTEL_PT_STATE_IN_SYNC,
61 INTEL_PT_STATE_TNT_CONT,
64 INTEL_PT_STATE_TIP_PGD,
66 INTEL_PT_STATE_FUP_NO_TIP,
69 static inline bool intel_pt_sample_time(enum intel_pt_pkt_state pkt_state)
72 case INTEL_PT_STATE_NO_PSB:
73 case INTEL_PT_STATE_NO_IP:
74 case INTEL_PT_STATE_ERR_RESYNC:
75 case INTEL_PT_STATE_IN_SYNC:
76 case INTEL_PT_STATE_TNT_CONT:
78 case INTEL_PT_STATE_TNT:
79 case INTEL_PT_STATE_TIP:
80 case INTEL_PT_STATE_TIP_PGD:
81 case INTEL_PT_STATE_FUP:
82 case INTEL_PT_STATE_FUP_NO_TIP:
89 #ifdef INTEL_PT_STRICT
90 #define INTEL_PT_STATE_ERR1 INTEL_PT_STATE_NO_PSB
91 #define INTEL_PT_STATE_ERR2 INTEL_PT_STATE_NO_PSB
92 #define INTEL_PT_STATE_ERR3 INTEL_PT_STATE_NO_PSB
93 #define INTEL_PT_STATE_ERR4 INTEL_PT_STATE_NO_PSB
95 #define INTEL_PT_STATE_ERR1 (decoder->pkt_state)
96 #define INTEL_PT_STATE_ERR2 INTEL_PT_STATE_NO_IP
97 #define INTEL_PT_STATE_ERR3 INTEL_PT_STATE_ERR_RESYNC
98 #define INTEL_PT_STATE_ERR4 INTEL_PT_STATE_IN_SYNC
101 struct intel_pt_decoder {
102 int (*get_trace)(struct intel_pt_buffer *buffer, void *data);
103 int (*walk_insn)(struct intel_pt_insn *intel_pt_insn,
104 uint64_t *insn_cnt_ptr, uint64_t *ip, uint64_t to_ip,
105 uint64_t max_insn_cnt, void *data);
107 struct intel_pt_state state;
108 const unsigned char *buf;
110 bool return_compression;
117 enum intel_pt_param_flags flags;
123 uint64_t tsc_timestamp;
124 uint64_t ref_timestamp;
125 uint64_t sample_timestamp;
127 uint64_t ctc_timestamp;
130 uint64_t cyc_ref_timestamp;
132 uint32_t tsc_ctc_ratio_n;
133 uint32_t tsc_ctc_ratio_d;
134 uint32_t tsc_ctc_mult;
136 uint32_t ctc_rem_mask;
138 struct intel_pt_stack stack;
139 enum intel_pt_pkt_state pkt_state;
140 struct intel_pt_pkt packet;
141 struct intel_pt_pkt tnt;
144 int last_packet_type;
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;
152 unsigned int insn_bytes;
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;
162 bool set_fup_tx_flags;
163 unsigned int fup_tx_flags;
164 unsigned int tx_flags;
165 uint64_t timestamp_insn_cnt;
166 uint64_t sample_insn_cnt;
171 const unsigned char *next_buf;
173 unsigned char temp_buf[INTEL_PT_PKT_MAX_SZ];
176 static uint64_t intel_pt_lower_power_of_2(uint64_t x)
180 for (i = 0; x != 1; i++)
186 static void intel_pt_setup_period(struct intel_pt_decoder *decoder)
188 if (decoder->period_type == INTEL_PT_PERIOD_TICKS) {
191 period = intel_pt_lower_power_of_2(decoder->period);
192 decoder->period_mask = ~(period - 1);
193 decoder->period_ticks = period;
197 static uint64_t multdiv(uint64_t t, uint32_t n, uint32_t d)
201 return (t / d) * n + ((t % d) * n) / d;
204 struct intel_pt_decoder *intel_pt_decoder_new(struct intel_pt_params *params)
206 struct intel_pt_decoder *decoder;
208 if (!params->get_trace || !params->walk_insn)
211 decoder = zalloc(sizeof(struct intel_pt_decoder));
215 decoder->get_trace = params->get_trace;
216 decoder->walk_insn = params->walk_insn;
217 decoder->data = params->data;
218 decoder->return_compression = params->return_compression;
220 decoder->flags = params->flags;
222 decoder->period = params->period;
223 decoder->period_type = params->period_type;
225 decoder->max_non_turbo_ratio = params->max_non_turbo_ratio;
226 decoder->max_non_turbo_ratio_fp = params->max_non_turbo_ratio;
228 intel_pt_setup_period(decoder);
230 decoder->mtc_shift = params->mtc_period;
231 decoder->ctc_rem_mask = (1 << decoder->mtc_shift) - 1;
233 decoder->tsc_ctc_ratio_n = params->tsc_ctc_ratio_n;
234 decoder->tsc_ctc_ratio_d = params->tsc_ctc_ratio_d;
236 if (!decoder->tsc_ctc_ratio_n)
237 decoder->tsc_ctc_ratio_d = 0;
239 if (decoder->tsc_ctc_ratio_d) {
240 if (!(decoder->tsc_ctc_ratio_n % decoder->tsc_ctc_ratio_d))
241 decoder->tsc_ctc_mult = decoder->tsc_ctc_ratio_n /
242 decoder->tsc_ctc_ratio_d;
246 * A TSC packet can slip past MTC packets so that the timestamp appears
247 * to go backwards. One estimate is that can be up to about 40 CPU
248 * cycles, which is certainly less than 0x1000 TSC ticks, but accept
249 * slippage an order of magnitude more to be on the safe side.
251 decoder->tsc_slip = 0x10000;
253 intel_pt_log("timestamp: mtc_shift %u\n", decoder->mtc_shift);
254 intel_pt_log("timestamp: tsc_ctc_ratio_n %u\n", decoder->tsc_ctc_ratio_n);
255 intel_pt_log("timestamp: tsc_ctc_ratio_d %u\n", decoder->tsc_ctc_ratio_d);
256 intel_pt_log("timestamp: tsc_ctc_mult %u\n", decoder->tsc_ctc_mult);
257 intel_pt_log("timestamp: tsc_slip %#x\n", decoder->tsc_slip);
262 static void intel_pt_pop_blk(struct intel_pt_stack *stack)
264 struct intel_pt_blk *blk = stack->blk;
266 stack->blk = blk->prev;
273 static uint64_t intel_pt_pop(struct intel_pt_stack *stack)
278 intel_pt_pop_blk(stack);
281 stack->pos = INTEL_PT_BLK_SIZE;
283 return stack->blk->ip[--stack->pos];
286 static int intel_pt_alloc_blk(struct intel_pt_stack *stack)
288 struct intel_pt_blk *blk;
294 blk = malloc(sizeof(struct intel_pt_blk));
299 blk->prev = stack->blk;
305 static int intel_pt_push(struct intel_pt_stack *stack, uint64_t ip)
309 if (!stack->blk || stack->pos == INTEL_PT_BLK_SIZE) {
310 err = intel_pt_alloc_blk(stack);
315 stack->blk->ip[stack->pos++] = ip;
319 static void intel_pt_clear_stack(struct intel_pt_stack *stack)
322 intel_pt_pop_blk(stack);
326 static void intel_pt_free_stack(struct intel_pt_stack *stack)
328 intel_pt_clear_stack(stack);
330 zfree(&stack->spare);
333 void intel_pt_decoder_free(struct intel_pt_decoder *decoder)
335 intel_pt_free_stack(&decoder->stack);
339 static int intel_pt_ext_err(int code)
343 return INTEL_PT_ERR_NOMEM;
345 return INTEL_PT_ERR_INTERN;
347 return INTEL_PT_ERR_BADPKT;
349 return INTEL_PT_ERR_NODATA;
351 return INTEL_PT_ERR_NOINSN;
353 return INTEL_PT_ERR_MISMAT;
355 return INTEL_PT_ERR_OVR;
357 return INTEL_PT_ERR_LOST;
359 return INTEL_PT_ERR_NELOOP;
361 return INTEL_PT_ERR_UNK;
365 static const char *intel_pt_err_msgs[] = {
366 [INTEL_PT_ERR_NOMEM] = "Memory allocation failed",
367 [INTEL_PT_ERR_INTERN] = "Internal error",
368 [INTEL_PT_ERR_BADPKT] = "Bad packet",
369 [INTEL_PT_ERR_NODATA] = "No more data",
370 [INTEL_PT_ERR_NOINSN] = "Failed to get instruction",
371 [INTEL_PT_ERR_MISMAT] = "Trace doesn't match instruction",
372 [INTEL_PT_ERR_OVR] = "Overflow packet",
373 [INTEL_PT_ERR_LOST] = "Lost trace data",
374 [INTEL_PT_ERR_UNK] = "Unknown error!",
375 [INTEL_PT_ERR_NELOOP] = "Never-ending loop",
378 int intel_pt__strerror(int code, char *buf, size_t buflen)
380 if (code < 1 || code > INTEL_PT_ERR_MAX)
381 code = INTEL_PT_ERR_UNK;
382 strlcpy(buf, intel_pt_err_msgs[code], buflen);
386 static uint64_t intel_pt_calc_ip(const struct intel_pt_pkt *packet,
391 switch (packet->count) {
393 ip = (last_ip & (uint64_t)0xffffffffffff0000ULL) |
397 ip = (last_ip & (uint64_t)0xffffffff00000000ULL) |
401 ip = packet->payload;
402 /* Sign-extend 6-byte ip */
403 if (ip & (uint64_t)0x800000000000ULL)
404 ip |= (uint64_t)0xffff000000000000ULL;
407 ip = (last_ip & (uint64_t)0xffff000000000000ULL) |
411 ip = packet->payload;
420 static inline void intel_pt_set_last_ip(struct intel_pt_decoder *decoder)
422 decoder->last_ip = intel_pt_calc_ip(&decoder->packet, decoder->last_ip);
423 decoder->have_last_ip = true;
426 static inline void intel_pt_set_ip(struct intel_pt_decoder *decoder)
428 intel_pt_set_last_ip(decoder);
429 decoder->ip = decoder->last_ip;
432 static void intel_pt_decoder_log_packet(struct intel_pt_decoder *decoder)
434 intel_pt_log_packet(&decoder->packet, decoder->pkt_len, decoder->pos,
438 static int intel_pt_bug(struct intel_pt_decoder *decoder)
440 intel_pt_log("ERROR: Internal error\n");
441 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
445 static inline void intel_pt_clear_tx_flags(struct intel_pt_decoder *decoder)
447 decoder->tx_flags = 0;
450 static inline void intel_pt_update_in_tx(struct intel_pt_decoder *decoder)
452 decoder->tx_flags = decoder->packet.payload & INTEL_PT_IN_TX;
455 static int intel_pt_bad_packet(struct intel_pt_decoder *decoder)
457 intel_pt_clear_tx_flags(decoder);
458 decoder->have_tma = false;
459 decoder->pkt_len = 1;
460 decoder->pkt_step = 1;
461 intel_pt_decoder_log_packet(decoder);
462 if (decoder->pkt_state != INTEL_PT_STATE_NO_PSB) {
463 intel_pt_log("ERROR: Bad packet\n");
464 decoder->pkt_state = INTEL_PT_STATE_ERR1;
469 static int intel_pt_get_data(struct intel_pt_decoder *decoder)
471 struct intel_pt_buffer buffer = { .buf = 0, };
474 decoder->pkt_step = 0;
476 intel_pt_log("Getting more data\n");
477 ret = decoder->get_trace(&buffer, decoder->data);
480 decoder->buf = buffer.buf;
481 decoder->len = buffer.len;
483 intel_pt_log("No more data\n");
486 if (!buffer.consecutive) {
488 decoder->pkt_state = INTEL_PT_STATE_NO_PSB;
489 decoder->ref_timestamp = buffer.ref_timestamp;
490 decoder->timestamp = 0;
491 decoder->have_tma = false;
492 decoder->state.trace_nr = buffer.trace_nr;
493 intel_pt_log("Reference timestamp 0x%" PRIx64 "\n",
494 decoder->ref_timestamp);
501 static int intel_pt_get_next_data(struct intel_pt_decoder *decoder)
503 if (!decoder->next_buf)
504 return intel_pt_get_data(decoder);
506 decoder->buf = decoder->next_buf;
507 decoder->len = decoder->next_len;
508 decoder->next_buf = 0;
509 decoder->next_len = 0;
513 static int intel_pt_get_split_packet(struct intel_pt_decoder *decoder)
515 unsigned char *buf = decoder->temp_buf;
516 size_t old_len, len, n;
519 old_len = decoder->len;
521 memcpy(buf, decoder->buf, len);
523 ret = intel_pt_get_data(decoder);
525 decoder->pos += old_len;
526 return ret < 0 ? ret : -EINVAL;
529 n = INTEL_PT_PKT_MAX_SZ - len;
530 if (n > decoder->len)
532 memcpy(buf + len, decoder->buf, n);
535 ret = intel_pt_get_packet(buf, len, &decoder->packet);
536 if (ret < (int)old_len) {
537 decoder->next_buf = decoder->buf;
538 decoder->next_len = decoder->len;
540 decoder->len = old_len;
541 return intel_pt_bad_packet(decoder);
544 decoder->next_buf = decoder->buf + (ret - old_len);
545 decoder->next_len = decoder->len - (ret - old_len);
553 struct intel_pt_pkt_info {
554 struct intel_pt_decoder *decoder;
555 struct intel_pt_pkt packet;
558 int last_packet_type;
562 typedef int (*intel_pt_pkt_cb_t)(struct intel_pt_pkt_info *pkt_info);
564 /* Lookahead packets in current buffer */
565 static int intel_pt_pkt_lookahead(struct intel_pt_decoder *decoder,
566 intel_pt_pkt_cb_t cb, void *data)
568 struct intel_pt_pkt_info pkt_info;
569 const unsigned char *buf = decoder->buf;
570 size_t len = decoder->len;
573 pkt_info.decoder = decoder;
574 pkt_info.pos = decoder->pos;
575 pkt_info.pkt_len = decoder->pkt_step;
576 pkt_info.last_packet_type = decoder->last_packet_type;
577 pkt_info.data = data;
581 pkt_info.pos += pkt_info.pkt_len;
582 buf += pkt_info.pkt_len;
583 len -= pkt_info.pkt_len;
586 return INTEL_PT_NEED_MORE_BYTES;
588 ret = intel_pt_get_packet(buf, len, &pkt_info.packet);
590 return INTEL_PT_NEED_MORE_BYTES;
594 pkt_info.pkt_len = ret;
595 } while (pkt_info.packet.type == INTEL_PT_PAD);
601 pkt_info.last_packet_type = pkt_info.packet.type;
605 struct intel_pt_calc_cyc_to_tsc_info {
609 uint64_t ctc_timestamp;
611 uint64_t tsc_timestamp;
616 double cbr_cyc_to_tsc;
620 * MTC provides a 8-bit slice of CTC but the TMA packet only provides the lower
621 * 16 bits of CTC. If mtc_shift > 8 then some of the MTC bits are not in the CTC
622 * provided by the TMA packet. Fix-up the last_mtc calculated from the TMA
623 * packet by copying the missing bits from the current MTC assuming the least
624 * difference between the two, and that the current MTC comes after last_mtc.
626 static void intel_pt_fixup_last_mtc(uint32_t mtc, int mtc_shift,
629 uint32_t first_missing_bit = 1U << (16 - mtc_shift);
630 uint32_t mask = ~(first_missing_bit - 1);
632 *last_mtc |= mtc & mask;
633 if (*last_mtc >= mtc) {
634 *last_mtc -= first_missing_bit;
639 static int intel_pt_calc_cyc_cb(struct intel_pt_pkt_info *pkt_info)
641 struct intel_pt_decoder *decoder = pkt_info->decoder;
642 struct intel_pt_calc_cyc_to_tsc_info *data = pkt_info->data;
646 uint32_t mtc, mtc_delta, ctc, fc, ctc_rem;
648 switch (pkt_info->packet.type) {
650 case INTEL_PT_TIP_PGE:
655 case INTEL_PT_MODE_EXEC:
656 case INTEL_PT_MODE_TSX:
657 case INTEL_PT_PSBEND:
667 mtc = pkt_info->packet.payload;
668 if (decoder->mtc_shift > 8 && data->fixup_last_mtc) {
669 data->fixup_last_mtc = false;
670 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
673 if (mtc > data->last_mtc)
674 mtc_delta = mtc - data->last_mtc;
676 mtc_delta = mtc + 256 - data->last_mtc;
677 data->ctc_delta += mtc_delta << decoder->mtc_shift;
678 data->last_mtc = mtc;
680 if (decoder->tsc_ctc_mult) {
681 timestamp = data->ctc_timestamp +
682 data->ctc_delta * decoder->tsc_ctc_mult;
684 timestamp = data->ctc_timestamp +
685 multdiv(data->ctc_delta,
686 decoder->tsc_ctc_ratio_n,
687 decoder->tsc_ctc_ratio_d);
690 if (timestamp < data->timestamp)
693 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
694 data->timestamp = timestamp;
701 timestamp = pkt_info->packet.payload |
702 (data->timestamp & (0xffULL << 56));
703 if (data->from_mtc && timestamp < data->timestamp &&
704 data->timestamp - timestamp < decoder->tsc_slip)
706 if (timestamp < data->timestamp)
707 timestamp += (1ULL << 56);
708 if (pkt_info->last_packet_type != INTEL_PT_CYC) {
711 data->tsc_timestamp = timestamp;
712 data->timestamp = timestamp;
721 if (!decoder->tsc_ctc_ratio_d)
724 ctc = pkt_info->packet.payload;
725 fc = pkt_info->packet.count;
726 ctc_rem = ctc & decoder->ctc_rem_mask;
728 data->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
730 data->ctc_timestamp = data->tsc_timestamp - fc;
731 if (decoder->tsc_ctc_mult) {
732 data->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
734 data->ctc_timestamp -=
735 multdiv(ctc_rem, decoder->tsc_ctc_ratio_n,
736 decoder->tsc_ctc_ratio_d);
740 data->have_tma = true;
741 data->fixup_last_mtc = true;
746 data->cycle_cnt += pkt_info->packet.payload;
750 cbr = pkt_info->packet.payload;
751 if (data->cbr && data->cbr != cbr)
754 data->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
757 case INTEL_PT_TIP_PGD:
758 case INTEL_PT_TRACESTOP:
760 case INTEL_PT_BAD: /* Does not happen */
765 if (!data->cbr && decoder->cbr) {
766 data->cbr = decoder->cbr;
767 data->cbr_cyc_to_tsc = decoder->cbr_cyc_to_tsc;
770 if (!data->cycle_cnt)
773 cyc_to_tsc = (double)(timestamp - decoder->timestamp) / data->cycle_cnt;
775 if (data->cbr && cyc_to_tsc > data->cbr_cyc_to_tsc &&
776 cyc_to_tsc / data->cbr_cyc_to_tsc > 1.25) {
777 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle too big (c.f. CBR-based value %g), pos " x64_fmt "\n",
778 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
782 decoder->calc_cyc_to_tsc = cyc_to_tsc;
783 decoder->have_calc_cyc_to_tsc = true;
786 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. CBR-based value %g, pos " x64_fmt "\n",
787 cyc_to_tsc, data->cbr_cyc_to_tsc, pkt_info->pos);
789 intel_pt_log("Timestamp: calculated %g TSC ticks per cycle c.f. unknown CBR-based value, pos " x64_fmt "\n",
790 cyc_to_tsc, pkt_info->pos);
796 static void intel_pt_calc_cyc_to_tsc(struct intel_pt_decoder *decoder,
799 struct intel_pt_calc_cyc_to_tsc_info data = {
802 .last_mtc = decoder->last_mtc,
803 .ctc_timestamp = decoder->ctc_timestamp,
804 .ctc_delta = decoder->ctc_delta,
805 .tsc_timestamp = decoder->tsc_timestamp,
806 .timestamp = decoder->timestamp,
807 .have_tma = decoder->have_tma,
808 .fixup_last_mtc = decoder->fixup_last_mtc,
809 .from_mtc = from_mtc,
813 intel_pt_pkt_lookahead(decoder, intel_pt_calc_cyc_cb, &data);
816 static int intel_pt_get_next_packet(struct intel_pt_decoder *decoder)
820 decoder->last_packet_type = decoder->packet.type;
823 decoder->pos += decoder->pkt_step;
824 decoder->buf += decoder->pkt_step;
825 decoder->len -= decoder->pkt_step;
828 ret = intel_pt_get_next_data(decoder);
833 ret = intel_pt_get_packet(decoder->buf, decoder->len,
835 if (ret == INTEL_PT_NEED_MORE_BYTES &&
836 decoder->len < INTEL_PT_PKT_MAX_SZ && !decoder->next_buf) {
837 ret = intel_pt_get_split_packet(decoder);
842 return intel_pt_bad_packet(decoder);
844 decoder->pkt_len = ret;
845 decoder->pkt_step = ret;
846 intel_pt_decoder_log_packet(decoder);
847 } while (decoder->packet.type == INTEL_PT_PAD);
852 static uint64_t intel_pt_next_period(struct intel_pt_decoder *decoder)
854 uint64_t timestamp, masked_timestamp;
856 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
857 masked_timestamp = timestamp & decoder->period_mask;
858 if (decoder->continuous_period) {
859 if (masked_timestamp > decoder->last_masked_timestamp)
863 masked_timestamp = timestamp & decoder->period_mask;
864 if (masked_timestamp > decoder->last_masked_timestamp) {
865 decoder->last_masked_timestamp = masked_timestamp;
866 decoder->continuous_period = true;
870 if (masked_timestamp < decoder->last_masked_timestamp)
871 return decoder->period_ticks;
873 return decoder->period_ticks - (timestamp - masked_timestamp);
876 static uint64_t intel_pt_next_sample(struct intel_pt_decoder *decoder)
878 switch (decoder->period_type) {
879 case INTEL_PT_PERIOD_INSTRUCTIONS:
880 return decoder->period - decoder->period_insn_cnt;
881 case INTEL_PT_PERIOD_TICKS:
882 return intel_pt_next_period(decoder);
883 case INTEL_PT_PERIOD_NONE:
884 case INTEL_PT_PERIOD_MTC:
890 static void intel_pt_sample_insn(struct intel_pt_decoder *decoder)
892 uint64_t timestamp, masked_timestamp;
894 switch (decoder->period_type) {
895 case INTEL_PT_PERIOD_INSTRUCTIONS:
896 decoder->period_insn_cnt = 0;
898 case INTEL_PT_PERIOD_TICKS:
899 timestamp = decoder->timestamp + decoder->timestamp_insn_cnt;
900 masked_timestamp = timestamp & decoder->period_mask;
901 if (masked_timestamp > decoder->last_masked_timestamp)
902 decoder->last_masked_timestamp = masked_timestamp;
904 decoder->last_masked_timestamp += decoder->period_ticks;
906 case INTEL_PT_PERIOD_NONE:
907 case INTEL_PT_PERIOD_MTC:
912 decoder->state.type |= INTEL_PT_INSTRUCTION;
915 static int intel_pt_walk_insn(struct intel_pt_decoder *decoder,
916 struct intel_pt_insn *intel_pt_insn, uint64_t ip)
918 uint64_t max_insn_cnt, insn_cnt = 0;
921 if (!decoder->mtc_insn)
922 decoder->mtc_insn = true;
924 max_insn_cnt = intel_pt_next_sample(decoder);
926 err = decoder->walk_insn(intel_pt_insn, &insn_cnt, &decoder->ip, ip,
927 max_insn_cnt, decoder->data);
929 decoder->tot_insn_cnt += insn_cnt;
930 decoder->timestamp_insn_cnt += insn_cnt;
931 decoder->sample_insn_cnt += insn_cnt;
932 decoder->period_insn_cnt += insn_cnt;
935 decoder->no_progress = 0;
936 decoder->pkt_state = INTEL_PT_STATE_ERR2;
937 intel_pt_log_at("ERROR: Failed to get instruction",
944 if (ip && decoder->ip == ip) {
949 if (max_insn_cnt && insn_cnt >= max_insn_cnt)
950 intel_pt_sample_insn(decoder);
952 if (intel_pt_insn->branch == INTEL_PT_BR_NO_BRANCH) {
953 decoder->state.type = INTEL_PT_INSTRUCTION;
954 decoder->state.from_ip = decoder->ip;
955 decoder->state.to_ip = 0;
956 decoder->ip += intel_pt_insn->length;
957 err = INTEL_PT_RETURN;
961 if (intel_pt_insn->op == INTEL_PT_OP_CALL) {
962 /* Zero-length calls are excluded */
963 if (intel_pt_insn->branch != INTEL_PT_BR_UNCONDITIONAL ||
964 intel_pt_insn->rel) {
965 err = intel_pt_push(&decoder->stack, decoder->ip +
966 intel_pt_insn->length);
970 } else if (intel_pt_insn->op == INTEL_PT_OP_RET) {
971 decoder->ret_addr = intel_pt_pop(&decoder->stack);
974 if (intel_pt_insn->branch == INTEL_PT_BR_UNCONDITIONAL) {
975 int cnt = decoder->no_progress++;
977 decoder->state.from_ip = decoder->ip;
978 decoder->ip += intel_pt_insn->length +
980 decoder->state.to_ip = decoder->ip;
981 err = INTEL_PT_RETURN;
984 * Check for being stuck in a loop. This can happen if a
985 * decoder error results in the decoder erroneously setting the
986 * ip to an address that is itself in an infinite loop that
987 * consumes no packets. When that happens, there must be an
988 * unconditional branch.
992 decoder->stuck_ip = decoder->state.to_ip;
993 decoder->stuck_ip_prd = 1;
994 decoder->stuck_ip_cnt = 1;
995 } else if (cnt > INTEL_PT_MAX_LOOPS ||
996 decoder->state.to_ip == decoder->stuck_ip) {
997 intel_pt_log_at("ERROR: Never-ending loop",
998 decoder->state.to_ip);
999 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1002 } else if (!--decoder->stuck_ip_cnt) {
1003 decoder->stuck_ip_prd += 1;
1004 decoder->stuck_ip_cnt = decoder->stuck_ip_prd;
1005 decoder->stuck_ip = decoder->state.to_ip;
1008 goto out_no_progress;
1011 decoder->no_progress = 0;
1013 decoder->state.insn_op = intel_pt_insn->op;
1014 decoder->state.insn_len = intel_pt_insn->length;
1016 if (decoder->tx_flags & INTEL_PT_IN_TX)
1017 decoder->state.flags |= INTEL_PT_IN_TX;
1022 static inline bool intel_pt_fup_with_nlip(struct intel_pt_decoder *decoder,
1023 struct intel_pt_insn *intel_pt_insn,
1024 uint64_t ip, int err)
1026 return decoder->flags & INTEL_PT_FUP_WITH_NLIP && !err &&
1027 intel_pt_insn->branch == INTEL_PT_BR_INDIRECT &&
1028 ip == decoder->ip + intel_pt_insn->length;
1031 static int intel_pt_walk_fup(struct intel_pt_decoder *decoder)
1033 struct intel_pt_insn intel_pt_insn;
1037 ip = decoder->last_ip;
1040 err = intel_pt_walk_insn(decoder, &intel_pt_insn, ip);
1041 if (err == INTEL_PT_RETURN)
1043 if (err == -EAGAIN ||
1044 intel_pt_fup_with_nlip(decoder, &intel_pt_insn, ip, err)) {
1045 if (decoder->set_fup_tx_flags) {
1046 decoder->set_fup_tx_flags = false;
1047 decoder->tx_flags = decoder->fup_tx_flags;
1048 decoder->state.type = INTEL_PT_TRANSACTION;
1049 decoder->state.from_ip = decoder->ip;
1050 decoder->state.to_ip = 0;
1051 decoder->state.flags = decoder->fup_tx_flags;
1056 decoder->set_fup_tx_flags = false;
1060 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1061 intel_pt_log_at("ERROR: Unexpected indirect branch",
1063 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1067 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1068 intel_pt_log_at("ERROR: Unexpected conditional branch",
1070 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1074 intel_pt_bug(decoder);
1078 static int intel_pt_walk_tip(struct intel_pt_decoder *decoder)
1080 struct intel_pt_insn intel_pt_insn;
1083 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1084 if (err == INTEL_PT_RETURN)
1089 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1090 if (decoder->pkt_state == INTEL_PT_STATE_TIP_PGD) {
1091 decoder->pge = false;
1092 decoder->continuous_period = false;
1093 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1094 decoder->state.from_ip = decoder->ip;
1095 decoder->state.to_ip = 0;
1096 if (decoder->packet.count != 0)
1097 decoder->ip = decoder->last_ip;
1099 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1100 decoder->state.from_ip = decoder->ip;
1101 if (decoder->packet.count == 0) {
1102 decoder->state.to_ip = 0;
1104 decoder->state.to_ip = decoder->last_ip;
1105 decoder->ip = decoder->last_ip;
1111 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1112 intel_pt_log_at("ERROR: Conditional branch when expecting indirect branch",
1114 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1118 return intel_pt_bug(decoder);
1121 static int intel_pt_walk_tnt(struct intel_pt_decoder *decoder)
1123 struct intel_pt_insn intel_pt_insn;
1127 err = intel_pt_walk_insn(decoder, &intel_pt_insn, 0);
1128 if (err == INTEL_PT_RETURN)
1133 if (intel_pt_insn.op == INTEL_PT_OP_RET) {
1134 if (!decoder->return_compression) {
1135 intel_pt_log_at("ERROR: RET when expecting conditional branch",
1137 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1140 if (!decoder->ret_addr) {
1141 intel_pt_log_at("ERROR: Bad RET compression (stack empty)",
1143 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1146 if (!(decoder->tnt.payload & BIT63)) {
1147 intel_pt_log_at("ERROR: Bad RET compression (TNT=N)",
1149 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1152 decoder->tnt.count -= 1;
1153 if (decoder->tnt.count)
1154 decoder->pkt_state = INTEL_PT_STATE_TNT_CONT;
1156 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1157 decoder->tnt.payload <<= 1;
1158 decoder->state.from_ip = decoder->ip;
1159 decoder->ip = decoder->ret_addr;
1160 decoder->state.to_ip = decoder->ip;
1164 if (intel_pt_insn.branch == INTEL_PT_BR_INDIRECT) {
1165 /* Handle deferred TIPs */
1166 err = intel_pt_get_next_packet(decoder);
1169 if (decoder->packet.type != INTEL_PT_TIP ||
1170 decoder->packet.count == 0) {
1171 intel_pt_log_at("ERROR: Missing deferred TIP for indirect branch",
1173 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1174 decoder->pkt_step = 0;
1177 intel_pt_set_last_ip(decoder);
1178 decoder->state.from_ip = decoder->ip;
1179 decoder->state.to_ip = decoder->last_ip;
1180 decoder->ip = decoder->last_ip;
1184 if (intel_pt_insn.branch == INTEL_PT_BR_CONDITIONAL) {
1185 decoder->tnt.count -= 1;
1186 if (decoder->tnt.count)
1187 decoder->pkt_state = INTEL_PT_STATE_TNT_CONT;
1189 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1190 if (decoder->tnt.payload & BIT63) {
1191 decoder->tnt.payload <<= 1;
1192 decoder->state.from_ip = decoder->ip;
1193 decoder->ip += intel_pt_insn.length +
1195 decoder->state.to_ip = decoder->ip;
1198 /* Instruction sample for a non-taken branch */
1199 if (decoder->state.type & INTEL_PT_INSTRUCTION) {
1200 decoder->tnt.payload <<= 1;
1201 decoder->state.type = INTEL_PT_INSTRUCTION;
1202 decoder->state.from_ip = decoder->ip;
1203 decoder->state.to_ip = 0;
1204 decoder->ip += intel_pt_insn.length;
1207 decoder->ip += intel_pt_insn.length;
1208 if (!decoder->tnt.count) {
1209 decoder->sample_timestamp = decoder->timestamp;
1210 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
1213 decoder->tnt.payload <<= 1;
1217 return intel_pt_bug(decoder);
1221 static int intel_pt_mode_tsx(struct intel_pt_decoder *decoder, bool *no_tip)
1223 unsigned int fup_tx_flags;
1226 fup_tx_flags = decoder->packet.payload &
1227 (INTEL_PT_IN_TX | INTEL_PT_ABORT_TX);
1228 err = intel_pt_get_next_packet(decoder);
1231 if (decoder->packet.type == INTEL_PT_FUP) {
1232 decoder->fup_tx_flags = fup_tx_flags;
1233 decoder->set_fup_tx_flags = true;
1234 if (!(decoder->fup_tx_flags & INTEL_PT_ABORT_TX))
1237 intel_pt_log_at("ERROR: Missing FUP after MODE.TSX",
1239 intel_pt_update_in_tx(decoder);
1244 static void intel_pt_calc_tsc_timestamp(struct intel_pt_decoder *decoder)
1248 decoder->have_tma = false;
1250 if (decoder->ref_timestamp) {
1251 timestamp = decoder->packet.payload |
1252 (decoder->ref_timestamp & (0xffULL << 56));
1253 if (timestamp < decoder->ref_timestamp) {
1254 if (decoder->ref_timestamp - timestamp > (1ULL << 55))
1255 timestamp += (1ULL << 56);
1257 if (timestamp - decoder->ref_timestamp > (1ULL << 55))
1258 timestamp -= (1ULL << 56);
1260 decoder->tsc_timestamp = timestamp;
1261 decoder->timestamp = timestamp;
1262 decoder->ref_timestamp = 0;
1263 decoder->timestamp_insn_cnt = 0;
1264 } else if (decoder->timestamp) {
1265 timestamp = decoder->packet.payload |
1266 (decoder->timestamp & (0xffULL << 56));
1267 decoder->tsc_timestamp = timestamp;
1268 if (timestamp < decoder->timestamp &&
1269 decoder->timestamp - timestamp < decoder->tsc_slip) {
1270 intel_pt_log_to("Suppressing backwards timestamp",
1272 timestamp = decoder->timestamp;
1274 if (timestamp < decoder->timestamp) {
1275 intel_pt_log_to("Wraparound timestamp", timestamp);
1276 timestamp += (1ULL << 56);
1277 decoder->tsc_timestamp = timestamp;
1279 decoder->timestamp = timestamp;
1280 decoder->timestamp_insn_cnt = 0;
1283 if (decoder->last_packet_type == INTEL_PT_CYC) {
1284 decoder->cyc_ref_timestamp = decoder->timestamp;
1285 decoder->cycle_cnt = 0;
1286 decoder->have_calc_cyc_to_tsc = false;
1287 intel_pt_calc_cyc_to_tsc(decoder, false);
1290 intel_pt_log_to("Setting timestamp", decoder->timestamp);
1293 static int intel_pt_overflow(struct intel_pt_decoder *decoder)
1295 intel_pt_log("ERROR: Buffer overflow\n");
1296 intel_pt_clear_tx_flags(decoder);
1297 decoder->timestamp_insn_cnt = 0;
1298 decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
1299 decoder->overflow = true;
1303 static void intel_pt_calc_tma(struct intel_pt_decoder *decoder)
1305 uint32_t ctc = decoder->packet.payload;
1306 uint32_t fc = decoder->packet.count;
1307 uint32_t ctc_rem = ctc & decoder->ctc_rem_mask;
1309 if (!decoder->tsc_ctc_ratio_d)
1312 decoder->last_mtc = (ctc >> decoder->mtc_shift) & 0xff;
1313 decoder->ctc_timestamp = decoder->tsc_timestamp - fc;
1314 if (decoder->tsc_ctc_mult) {
1315 decoder->ctc_timestamp -= ctc_rem * decoder->tsc_ctc_mult;
1317 decoder->ctc_timestamp -= multdiv(ctc_rem,
1318 decoder->tsc_ctc_ratio_n,
1319 decoder->tsc_ctc_ratio_d);
1321 decoder->ctc_delta = 0;
1322 decoder->have_tma = true;
1323 decoder->fixup_last_mtc = true;
1324 intel_pt_log("CTC timestamp " x64_fmt " last MTC %#x CTC rem %#x\n",
1325 decoder->ctc_timestamp, decoder->last_mtc, ctc_rem);
1328 static void intel_pt_calc_mtc_timestamp(struct intel_pt_decoder *decoder)
1331 uint32_t mtc, mtc_delta;
1333 if (!decoder->have_tma)
1336 mtc = decoder->packet.payload;
1338 if (decoder->mtc_shift > 8 && decoder->fixup_last_mtc) {
1339 decoder->fixup_last_mtc = false;
1340 intel_pt_fixup_last_mtc(mtc, decoder->mtc_shift,
1341 &decoder->last_mtc);
1344 if (mtc > decoder->last_mtc)
1345 mtc_delta = mtc - decoder->last_mtc;
1347 mtc_delta = mtc + 256 - decoder->last_mtc;
1349 decoder->ctc_delta += mtc_delta << decoder->mtc_shift;
1351 if (decoder->tsc_ctc_mult) {
1352 timestamp = decoder->ctc_timestamp +
1353 decoder->ctc_delta * decoder->tsc_ctc_mult;
1355 timestamp = decoder->ctc_timestamp +
1356 multdiv(decoder->ctc_delta,
1357 decoder->tsc_ctc_ratio_n,
1358 decoder->tsc_ctc_ratio_d);
1361 if (timestamp < decoder->timestamp)
1362 intel_pt_log("Suppressing MTC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1363 timestamp, decoder->timestamp);
1365 decoder->timestamp = timestamp;
1367 decoder->timestamp_insn_cnt = 0;
1368 decoder->last_mtc = mtc;
1370 if (decoder->last_packet_type == INTEL_PT_CYC) {
1371 decoder->cyc_ref_timestamp = decoder->timestamp;
1372 decoder->cycle_cnt = 0;
1373 decoder->have_calc_cyc_to_tsc = false;
1374 intel_pt_calc_cyc_to_tsc(decoder, true);
1378 static void intel_pt_calc_cbr(struct intel_pt_decoder *decoder)
1380 unsigned int cbr = decoder->packet.payload;
1382 if (decoder->cbr == cbr)
1386 decoder->cbr_cyc_to_tsc = decoder->max_non_turbo_ratio_fp / cbr;
1389 static void intel_pt_calc_cyc_timestamp(struct intel_pt_decoder *decoder)
1391 uint64_t timestamp = decoder->cyc_ref_timestamp;
1393 decoder->have_cyc = true;
1395 decoder->cycle_cnt += decoder->packet.payload;
1397 if (!decoder->cyc_ref_timestamp)
1400 if (decoder->have_calc_cyc_to_tsc)
1401 timestamp += decoder->cycle_cnt * decoder->calc_cyc_to_tsc;
1402 else if (decoder->cbr)
1403 timestamp += decoder->cycle_cnt * decoder->cbr_cyc_to_tsc;
1407 if (timestamp < decoder->timestamp)
1408 intel_pt_log("Suppressing CYC timestamp " x64_fmt " less than current timestamp " x64_fmt "\n",
1409 timestamp, decoder->timestamp);
1411 decoder->timestamp = timestamp;
1413 decoder->timestamp_insn_cnt = 0;
1416 /* Walk PSB+ packets when already in sync. */
1417 static int intel_pt_walk_psbend(struct intel_pt_decoder *decoder)
1422 err = intel_pt_get_next_packet(decoder);
1426 switch (decoder->packet.type) {
1427 case INTEL_PT_PSBEND:
1430 case INTEL_PT_TIP_PGD:
1431 case INTEL_PT_TIP_PGE:
1434 case INTEL_PT_TRACESTOP:
1437 decoder->have_tma = false;
1438 intel_pt_log("ERROR: Unexpected packet\n");
1442 return intel_pt_overflow(decoder);
1445 intel_pt_calc_tsc_timestamp(decoder);
1449 intel_pt_calc_tma(decoder);
1453 intel_pt_calc_cbr(decoder);
1456 case INTEL_PT_MODE_EXEC:
1457 decoder->exec_mode = decoder->packet.payload;
1461 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1465 decoder->pge = true;
1466 if (decoder->packet.count)
1467 intel_pt_set_last_ip(decoder);
1470 case INTEL_PT_MODE_TSX:
1471 intel_pt_update_in_tx(decoder);
1475 intel_pt_calc_mtc_timestamp(decoder);
1476 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1477 decoder->state.type |= INTEL_PT_INSTRUCTION;
1481 intel_pt_calc_cyc_timestamp(decoder);
1493 static int intel_pt_walk_fup_tip(struct intel_pt_decoder *decoder)
1497 if (decoder->tx_flags & INTEL_PT_ABORT_TX) {
1498 decoder->tx_flags = 0;
1499 decoder->state.flags &= ~INTEL_PT_IN_TX;
1500 decoder->state.flags |= INTEL_PT_ABORT_TX;
1502 decoder->state.flags |= INTEL_PT_ASYNC;
1506 err = intel_pt_get_next_packet(decoder);
1510 switch (decoder->packet.type) {
1513 case INTEL_PT_TRACESTOP:
1517 case INTEL_PT_MODE_TSX:
1519 case INTEL_PT_PSBEND:
1520 intel_pt_log("ERROR: Missing TIP after FUP\n");
1521 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1522 decoder->pkt_step = 0;
1526 intel_pt_calc_cbr(decoder);
1530 return intel_pt_overflow(decoder);
1532 case INTEL_PT_TIP_PGD:
1533 decoder->state.from_ip = decoder->ip;
1534 decoder->state.to_ip = 0;
1535 if (decoder->packet.count != 0) {
1536 intel_pt_set_ip(decoder);
1537 intel_pt_log("Omitting PGD ip " x64_fmt "\n",
1540 decoder->pge = false;
1541 decoder->continuous_period = false;
1544 case INTEL_PT_TIP_PGE:
1545 decoder->pge = true;
1546 intel_pt_log("Omitting PGE ip " x64_fmt "\n",
1548 decoder->state.from_ip = 0;
1549 if (decoder->packet.count == 0) {
1550 decoder->state.to_ip = 0;
1552 intel_pt_set_ip(decoder);
1553 decoder->state.to_ip = decoder->ip;
1558 decoder->state.from_ip = decoder->ip;
1559 if (decoder->packet.count == 0) {
1560 decoder->state.to_ip = 0;
1562 intel_pt_set_ip(decoder);
1563 decoder->state.to_ip = decoder->ip;
1568 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1572 intel_pt_calc_mtc_timestamp(decoder);
1573 if (decoder->period_type == INTEL_PT_PERIOD_MTC)
1574 decoder->state.type |= INTEL_PT_INSTRUCTION;
1578 intel_pt_calc_cyc_timestamp(decoder);
1581 case INTEL_PT_MODE_EXEC:
1582 decoder->exec_mode = decoder->packet.payload;
1591 return intel_pt_bug(decoder);
1596 static int intel_pt_walk_trace(struct intel_pt_decoder *decoder)
1598 bool no_tip = false;
1602 err = intel_pt_get_next_packet(decoder);
1606 switch (decoder->packet.type) {
1608 if (!decoder->packet.count)
1610 decoder->tnt = decoder->packet;
1611 decoder->pkt_state = INTEL_PT_STATE_TNT;
1612 err = intel_pt_walk_tnt(decoder);
1617 case INTEL_PT_TIP_PGD:
1618 if (decoder->packet.count != 0)
1619 intel_pt_set_last_ip(decoder);
1620 decoder->pkt_state = INTEL_PT_STATE_TIP_PGD;
1621 return intel_pt_walk_tip(decoder);
1623 case INTEL_PT_TIP_PGE: {
1624 decoder->pge = true;
1625 if (decoder->packet.count == 0) {
1626 intel_pt_log_at("Skipping zero TIP.PGE",
1630 intel_pt_set_ip(decoder);
1631 decoder->state.from_ip = 0;
1632 decoder->state.to_ip = decoder->ip;
1637 return intel_pt_overflow(decoder);
1640 if (decoder->packet.count != 0)
1641 intel_pt_set_last_ip(decoder);
1642 decoder->pkt_state = INTEL_PT_STATE_TIP;
1643 return intel_pt_walk_tip(decoder);
1646 if (decoder->packet.count == 0) {
1647 intel_pt_log_at("Skipping zero FUP",
1652 intel_pt_set_last_ip(decoder);
1653 err = intel_pt_walk_fup(decoder);
1654 if (err != -EAGAIN) {
1658 decoder->pkt_state =
1659 INTEL_PT_STATE_FUP_NO_TIP;
1661 decoder->pkt_state = INTEL_PT_STATE_FUP;
1668 return intel_pt_walk_fup_tip(decoder);
1670 case INTEL_PT_TRACESTOP:
1671 decoder->pge = false;
1672 decoder->continuous_period = false;
1673 intel_pt_clear_tx_flags(decoder);
1674 decoder->have_tma = false;
1678 decoder->last_ip = 0;
1679 decoder->have_last_ip = true;
1680 intel_pt_clear_stack(&decoder->stack);
1681 err = intel_pt_walk_psbend(decoder);
1689 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1693 intel_pt_calc_mtc_timestamp(decoder);
1694 if (decoder->period_type != INTEL_PT_PERIOD_MTC)
1697 * Ensure that there has been an instruction since the
1700 if (!decoder->mtc_insn)
1702 decoder->mtc_insn = false;
1703 /* Ensure that there is a timestamp */
1704 if (!decoder->timestamp)
1706 decoder->state.type = INTEL_PT_INSTRUCTION;
1707 decoder->state.from_ip = decoder->ip;
1708 decoder->state.to_ip = 0;
1709 decoder->mtc_insn = false;
1713 intel_pt_calc_tsc_timestamp(decoder);
1717 intel_pt_calc_tma(decoder);
1721 intel_pt_calc_cyc_timestamp(decoder);
1725 intel_pt_calc_cbr(decoder);
1728 case INTEL_PT_MODE_EXEC:
1729 decoder->exec_mode = decoder->packet.payload;
1732 case INTEL_PT_MODE_TSX:
1733 /* MODE_TSX need not be followed by FUP */
1734 if (!decoder->pge) {
1735 intel_pt_update_in_tx(decoder);
1738 err = intel_pt_mode_tsx(decoder, &no_tip);
1743 case INTEL_PT_BAD: /* Does not happen */
1744 return intel_pt_bug(decoder);
1746 case INTEL_PT_PSBEND:
1753 return intel_pt_bug(decoder);
1758 static inline bool intel_pt_have_ip(struct intel_pt_decoder *decoder)
1760 return decoder->packet.count &&
1761 (decoder->have_last_ip || decoder->packet.count == 3 ||
1762 decoder->packet.count == 6);
1765 /* Walk PSB+ packets to get in sync. */
1766 static int intel_pt_walk_psb(struct intel_pt_decoder *decoder)
1771 err = intel_pt_get_next_packet(decoder);
1775 switch (decoder->packet.type) {
1776 case INTEL_PT_TIP_PGD:
1777 decoder->continuous_period = false;
1779 case INTEL_PT_TIP_PGE:
1781 intel_pt_log("ERROR: Unexpected packet\n");
1785 decoder->pge = true;
1786 if (intel_pt_have_ip(decoder)) {
1787 uint64_t current_ip = decoder->ip;
1789 intel_pt_set_ip(decoder);
1791 intel_pt_log_to("Setting IP",
1797 intel_pt_calc_mtc_timestamp(decoder);
1801 intel_pt_calc_tsc_timestamp(decoder);
1805 intel_pt_calc_tma(decoder);
1809 intel_pt_calc_cyc_timestamp(decoder);
1813 intel_pt_calc_cbr(decoder);
1817 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1820 case INTEL_PT_MODE_EXEC:
1821 decoder->exec_mode = decoder->packet.payload;
1824 case INTEL_PT_MODE_TSX:
1825 intel_pt_update_in_tx(decoder);
1828 case INTEL_PT_TRACESTOP:
1829 decoder->pge = false;
1830 decoder->continuous_period = false;
1831 intel_pt_clear_tx_flags(decoder);
1835 decoder->have_tma = false;
1836 intel_pt_log("ERROR: Unexpected packet\n");
1838 decoder->pkt_state = INTEL_PT_STATE_ERR4;
1840 decoder->pkt_state = INTEL_PT_STATE_ERR3;
1843 case INTEL_PT_BAD: /* Does not happen */
1844 return intel_pt_bug(decoder);
1847 return intel_pt_overflow(decoder);
1849 case INTEL_PT_PSBEND:
1862 static int intel_pt_walk_to_ip(struct intel_pt_decoder *decoder)
1867 err = intel_pt_get_next_packet(decoder);
1871 switch (decoder->packet.type) {
1872 case INTEL_PT_TIP_PGD:
1873 decoder->continuous_period = false;
1875 case INTEL_PT_TIP_PGE:
1877 decoder->pge = decoder->packet.type != INTEL_PT_TIP_PGD;
1878 if (intel_pt_have_ip(decoder))
1879 intel_pt_set_ip(decoder);
1885 if (intel_pt_have_ip(decoder))
1886 intel_pt_set_ip(decoder);
1892 intel_pt_calc_mtc_timestamp(decoder);
1896 intel_pt_calc_tsc_timestamp(decoder);
1900 intel_pt_calc_tma(decoder);
1904 intel_pt_calc_cyc_timestamp(decoder);
1908 intel_pt_calc_cbr(decoder);
1912 decoder->cr3 = decoder->packet.payload & (BIT63 - 1);
1915 case INTEL_PT_MODE_EXEC:
1916 decoder->exec_mode = decoder->packet.payload;
1919 case INTEL_PT_MODE_TSX:
1920 intel_pt_update_in_tx(decoder);
1924 return intel_pt_overflow(decoder);
1926 case INTEL_PT_BAD: /* Does not happen */
1927 return intel_pt_bug(decoder);
1929 case INTEL_PT_TRACESTOP:
1930 decoder->pge = false;
1931 decoder->continuous_period = false;
1932 intel_pt_clear_tx_flags(decoder);
1933 decoder->have_tma = false;
1937 decoder->last_ip = 0;
1938 decoder->have_last_ip = true;
1939 intel_pt_clear_stack(&decoder->stack);
1940 err = intel_pt_walk_psb(decoder);
1944 /* Do not have a sample */
1945 decoder->state.type = 0;
1951 case INTEL_PT_PSBEND:
1961 static int intel_pt_sync_ip(struct intel_pt_decoder *decoder)
1965 decoder->set_fup_tx_flags = false;
1967 intel_pt_log("Scanning for full IP\n");
1968 err = intel_pt_walk_to_ip(decoder);
1972 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
1973 decoder->overflow = false;
1975 decoder->state.from_ip = 0;
1976 decoder->state.to_ip = decoder->ip;
1977 intel_pt_log_to("Setting IP", decoder->ip);
1982 static int intel_pt_part_psb(struct intel_pt_decoder *decoder)
1984 const unsigned char *end = decoder->buf + decoder->len;
1987 for (i = INTEL_PT_PSB_LEN - 1; i; i--) {
1988 if (i > decoder->len)
1990 if (!memcmp(end - i, INTEL_PT_PSB_STR, i))
1996 static int intel_pt_rest_psb(struct intel_pt_decoder *decoder, int part_psb)
1998 size_t rest_psb = INTEL_PT_PSB_LEN - part_psb;
1999 const char *psb = INTEL_PT_PSB_STR;
2001 if (rest_psb > decoder->len ||
2002 memcmp(decoder->buf, psb + part_psb, rest_psb))
2008 static int intel_pt_get_split_psb(struct intel_pt_decoder *decoder,
2013 decoder->pos += decoder->len;
2016 ret = intel_pt_get_next_data(decoder);
2020 rest_psb = intel_pt_rest_psb(decoder, part_psb);
2024 decoder->pos -= part_psb;
2025 decoder->next_buf = decoder->buf + rest_psb;
2026 decoder->next_len = decoder->len - rest_psb;
2027 memcpy(decoder->temp_buf, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2028 decoder->buf = decoder->temp_buf;
2029 decoder->len = INTEL_PT_PSB_LEN;
2034 static int intel_pt_scan_for_psb(struct intel_pt_decoder *decoder)
2036 unsigned char *next;
2039 intel_pt_log("Scanning for PSB\n");
2041 if (!decoder->len) {
2042 ret = intel_pt_get_next_data(decoder);
2047 next = memmem(decoder->buf, decoder->len, INTEL_PT_PSB_STR,
2052 part_psb = intel_pt_part_psb(decoder);
2054 ret = intel_pt_get_split_psb(decoder, part_psb);
2058 decoder->pos += decoder->len;
2064 decoder->pkt_step = next - decoder->buf;
2065 return intel_pt_get_next_packet(decoder);
2069 static int intel_pt_sync(struct intel_pt_decoder *decoder)
2073 decoder->pge = false;
2074 decoder->continuous_period = false;
2075 decoder->have_last_ip = false;
2076 decoder->last_ip = 0;
2078 intel_pt_clear_stack(&decoder->stack);
2080 err = intel_pt_scan_for_psb(decoder);
2084 decoder->have_last_ip = true;
2085 decoder->pkt_state = INTEL_PT_STATE_NO_IP;
2087 err = intel_pt_walk_psb(decoder);
2092 decoder->state.type = 0; /* Do not have a sample */
2093 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2095 return intel_pt_sync_ip(decoder);
2101 static uint64_t intel_pt_est_timestamp(struct intel_pt_decoder *decoder)
2103 uint64_t est = decoder->sample_insn_cnt << 1;
2105 if (!decoder->cbr || !decoder->max_non_turbo_ratio)
2108 est *= decoder->max_non_turbo_ratio;
2109 est /= decoder->cbr;
2111 return decoder->sample_timestamp + est;
2114 const struct intel_pt_state *intel_pt_decode(struct intel_pt_decoder *decoder)
2119 decoder->state.type = INTEL_PT_BRANCH;
2120 decoder->state.flags = 0;
2122 switch (decoder->pkt_state) {
2123 case INTEL_PT_STATE_NO_PSB:
2124 err = intel_pt_sync(decoder);
2126 case INTEL_PT_STATE_NO_IP:
2127 decoder->have_last_ip = false;
2128 decoder->last_ip = 0;
2131 case INTEL_PT_STATE_ERR_RESYNC:
2132 err = intel_pt_sync_ip(decoder);
2134 case INTEL_PT_STATE_IN_SYNC:
2135 err = intel_pt_walk_trace(decoder);
2137 case INTEL_PT_STATE_TNT:
2138 case INTEL_PT_STATE_TNT_CONT:
2139 err = intel_pt_walk_tnt(decoder);
2141 err = intel_pt_walk_trace(decoder);
2143 case INTEL_PT_STATE_TIP:
2144 case INTEL_PT_STATE_TIP_PGD:
2145 err = intel_pt_walk_tip(decoder);
2147 case INTEL_PT_STATE_FUP:
2148 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2149 err = intel_pt_walk_fup(decoder);
2151 err = intel_pt_walk_fup_tip(decoder);
2153 decoder->pkt_state = INTEL_PT_STATE_FUP;
2155 case INTEL_PT_STATE_FUP_NO_TIP:
2156 decoder->pkt_state = INTEL_PT_STATE_IN_SYNC;
2157 err = intel_pt_walk_fup(decoder);
2159 err = intel_pt_walk_trace(decoder);
2162 err = intel_pt_bug(decoder);
2165 } while (err == -ENOLINK);
2168 decoder->state.err = intel_pt_ext_err(err);
2169 decoder->state.from_ip = decoder->ip;
2170 decoder->sample_timestamp = decoder->timestamp;
2171 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
2173 decoder->state.err = 0;
2174 if (intel_pt_sample_time(decoder->pkt_state)) {
2175 decoder->sample_timestamp = decoder->timestamp;
2176 decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
2180 decoder->state.timestamp = decoder->sample_timestamp;
2181 decoder->state.est_timestamp = intel_pt_est_timestamp(decoder);
2182 decoder->state.cr3 = decoder->cr3;
2183 decoder->state.tot_insn_cnt = decoder->tot_insn_cnt;
2185 return &decoder->state;
2189 * intel_pt_next_psb - move buffer pointer to the start of the next PSB packet.
2190 * @buf: pointer to buffer pointer
2191 * @len: size of buffer
2193 * Updates the buffer pointer to point to the start of the next PSB packet if
2194 * there is one, otherwise the buffer pointer is unchanged. If @buf is updated,
2195 * @len is adjusted accordingly.
2197 * Return: %true if a PSB packet is found, %false otherwise.
2199 static bool intel_pt_next_psb(unsigned char **buf, size_t *len)
2201 unsigned char *next;
2203 next = memmem(*buf, *len, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2205 *len -= next - *buf;
2213 * intel_pt_step_psb - move buffer pointer to the start of the following PSB
2215 * @buf: pointer to buffer pointer
2216 * @len: size of buffer
2218 * Updates the buffer pointer to point to the start of the following PSB packet
2219 * (skipping the PSB at @buf itself) if there is one, otherwise the buffer
2220 * pointer is unchanged. If @buf is updated, @len is adjusted accordingly.
2222 * Return: %true if a PSB packet is found, %false otherwise.
2224 static bool intel_pt_step_psb(unsigned char **buf, size_t *len)
2226 unsigned char *next;
2231 next = memmem(*buf + 1, *len - 1, INTEL_PT_PSB_STR, INTEL_PT_PSB_LEN);
2233 *len -= next - *buf;
2241 * intel_pt_last_psb - find the last PSB packet in a buffer.
2243 * @len: size of buffer
2245 * This function finds the last PSB in a buffer.
2247 * Return: A pointer to the last PSB in @buf if found, %NULL otherwise.
2249 static unsigned char *intel_pt_last_psb(unsigned char *buf, size_t len)
2251 const char *n = INTEL_PT_PSB_STR;
2255 if (len < INTEL_PT_PSB_LEN)
2258 k = len - INTEL_PT_PSB_LEN + 1;
2260 p = memrchr(buf, n[0], k);
2263 if (!memcmp(p + 1, n + 1, INTEL_PT_PSB_LEN - 1))
2272 * intel_pt_next_tsc - find and return next TSC.
2274 * @len: size of buffer
2275 * @tsc: TSC value returned
2276 * @rem: returns remaining size when TSC is found
2278 * Find a TSC packet in @buf and return the TSC value. This function assumes
2279 * that @buf starts at a PSB and that PSB+ will contain TSC and so stops if a
2280 * PSBEND packet is found.
2282 * Return: %true if TSC is found, false otherwise.
2284 static bool intel_pt_next_tsc(unsigned char *buf, size_t len, uint64_t *tsc,
2287 struct intel_pt_pkt packet;
2291 ret = intel_pt_get_packet(buf, len, &packet);
2294 if (packet.type == INTEL_PT_TSC) {
2295 *tsc = packet.payload;
2299 if (packet.type == INTEL_PT_PSBEND)
2308 * intel_pt_tsc_cmp - compare 7-byte TSCs.
2309 * @tsc1: first TSC to compare
2310 * @tsc2: second TSC to compare
2312 * This function compares 7-byte TSC values allowing for the possibility that
2313 * TSC wrapped around. Generally it is not possible to know if TSC has wrapped
2314 * around so for that purpose this function assumes the absolute difference is
2315 * less than half the maximum difference.
2317 * Return: %-1 if @tsc1 is before @tsc2, %0 if @tsc1 == @tsc2, %1 if @tsc1 is
2320 static int intel_pt_tsc_cmp(uint64_t tsc1, uint64_t tsc2)
2322 const uint64_t halfway = (1ULL << 55);
2328 if (tsc2 - tsc1 < halfway)
2333 if (tsc1 - tsc2 < halfway)
2340 #define MAX_PADDING (PERF_AUXTRACE_RECORD_ALIGNMENT - 1)
2343 * adj_for_padding - adjust overlap to account for padding.
2344 * @buf_b: second buffer
2345 * @buf_a: first buffer
2346 * @len_a: size of first buffer
2348 * @buf_a might have up to 7 bytes of padding appended. Adjust the overlap
2351 * Return: A pointer into @buf_b from where non-overlapped data starts
2353 static unsigned char *adj_for_padding(unsigned char *buf_b,
2354 unsigned char *buf_a, size_t len_a)
2356 unsigned char *p = buf_b - MAX_PADDING;
2357 unsigned char *q = buf_a + len_a - MAX_PADDING;
2360 for (i = MAX_PADDING; i; i--, p++, q++) {
2369 * intel_pt_find_overlap_tsc - determine start of non-overlapped trace data
2371 * @buf_a: first buffer
2372 * @len_a: size of first buffer
2373 * @buf_b: second buffer
2374 * @len_b: size of second buffer
2375 * @consecutive: returns true if there is data in buf_b that is consecutive
2378 * If the trace contains TSC we can look at the last TSC of @buf_a and the
2379 * first TSC of @buf_b in order to determine if the buffers overlap, and then
2380 * walk forward in @buf_b until a later TSC is found. A precondition is that
2381 * @buf_a and @buf_b are positioned at a PSB.
2383 * Return: A pointer into @buf_b from where non-overlapped data starts, or
2384 * @buf_b + @len_b if there is no non-overlapped data.
2386 static unsigned char *intel_pt_find_overlap_tsc(unsigned char *buf_a,
2388 unsigned char *buf_b,
2389 size_t len_b, bool *consecutive)
2391 uint64_t tsc_a, tsc_b;
2393 size_t len, rem_a, rem_b;
2395 p = intel_pt_last_psb(buf_a, len_a);
2397 return buf_b; /* No PSB in buf_a => no overlap */
2399 len = len_a - (p - buf_a);
2400 if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a)) {
2401 /* The last PSB+ in buf_a is incomplete, so go back one more */
2403 p = intel_pt_last_psb(buf_a, len_a);
2405 return buf_b; /* No full PSB+ => assume no overlap */
2406 len = len_a - (p - buf_a);
2407 if (!intel_pt_next_tsc(p, len, &tsc_a, &rem_a))
2408 return buf_b; /* No TSC in buf_a => assume no overlap */
2412 /* Ignore PSB+ with no TSC */
2413 if (intel_pt_next_tsc(buf_b, len_b, &tsc_b, &rem_b)) {
2414 int cmp = intel_pt_tsc_cmp(tsc_a, tsc_b);
2416 /* Same TSC, so buffers are consecutive */
2417 if (!cmp && rem_b >= rem_a) {
2418 unsigned char *start;
2420 *consecutive = true;
2421 start = buf_b + len_b - (rem_b - rem_a);
2422 return adj_for_padding(start, buf_a, len_a);
2425 return buf_b; /* tsc_a < tsc_b => no overlap */
2428 if (!intel_pt_step_psb(&buf_b, &len_b))
2429 return buf_b + len_b; /* No PSB in buf_b => no data */
2434 * intel_pt_find_overlap - determine start of non-overlapped trace data.
2435 * @buf_a: first buffer
2436 * @len_a: size of first buffer
2437 * @buf_b: second buffer
2438 * @len_b: size of second buffer
2439 * @have_tsc: can use TSC packets to detect overlap
2440 * @consecutive: returns true if there is data in buf_b that is consecutive
2443 * When trace samples or snapshots are recorded there is the possibility that
2444 * the data overlaps. Note that, for the purposes of decoding, data is only
2445 * useful if it begins with a PSB packet.
2447 * Return: A pointer into @buf_b from where non-overlapped data starts, or
2448 * @buf_b + @len_b if there is no non-overlapped data.
2450 unsigned char *intel_pt_find_overlap(unsigned char *buf_a, size_t len_a,
2451 unsigned char *buf_b, size_t len_b,
2452 bool have_tsc, bool *consecutive)
2454 unsigned char *found;
2456 /* Buffer 'b' must start at PSB so throw away everything before that */
2457 if (!intel_pt_next_psb(&buf_b, &len_b))
2458 return buf_b + len_b; /* No PSB */
2460 if (!intel_pt_next_psb(&buf_a, &len_a))
2461 return buf_b; /* No overlap */
2464 found = intel_pt_find_overlap_tsc(buf_a, len_a, buf_b, len_b,
2471 * Buffer 'b' cannot end within buffer 'a' so, for comparison purposes,
2472 * we can ignore the first part of buffer 'a'.
2474 while (len_b < len_a) {
2475 if (!intel_pt_step_psb(&buf_a, &len_a))
2476 return buf_b; /* No overlap */
2479 /* Now len_b >= len_a */
2481 /* Potential overlap so check the bytes */
2482 found = memmem(buf_a, len_a, buf_b, len_a);
2484 *consecutive = true;
2485 return adj_for_padding(buf_b + len_a, buf_a, len_a);
2488 /* Try again at next PSB in buffer 'a' */
2489 if (!intel_pt_step_psb(&buf_a, &len_a))
2490 return buf_b; /* No overlap */