1 files changed, 288 insertions, 8 deletions

diff --git a/tools/perf/util/intel-pt.c b/tools/perf/util/intel-pt.c
index 8ed51f4e9e304..550db6e779685 100644
--- a/tools/perf/util/intel-pt.c
+++ b/tools/perf/util/intel-pt.c
@@ -35,6 +35,8 @@
 #include "config.h"
 #include "time-utils.h"
 
+#include "../arch/x86/include/uapi/asm/perf_regs.h"
+
 #include "intel-pt-decoder/intel-pt-log.h"
 #include "intel-pt-decoder/intel-pt-decoder.h"
 #include "intel-pt-decoder/intel-pt-insn-decoder.h"
@@ -101,6 +103,9 @@ struct intel_pt {
 	u64 pwrx_id;
 	u64 cbr_id;
 
+	bool sample_pebs;
+	struct perf_evsel *pebs_evsel;
+
 	u64 tsc_bit;
 	u64 mtc_bit;
 	u64 mtc_freq_bits;
@@ -177,13 +182,14 @@ static void intel_pt_dump(struct intel_pt *pt __maybe_unused,
 	int ret, pkt_len, i;
 	char desc[INTEL_PT_PKT_DESC_MAX];
 	const char *color = PERF_COLOR_BLUE;
+	enum intel_pt_pkt_ctx ctx = INTEL_PT_NO_CTX;
 
 	color_fprintf(stdout, color,
 		      ". ... Intel Processor Trace data: size %zu bytes\n",
 		      len);
 
 	while (len) {
-		ret = intel_pt_get_packet(buf, len, &packet);
+		ret = intel_pt_get_packet(buf, len, &packet, &ctx);
 		if (ret > 0)
 			pkt_len = ret;
 		else
@@ -1178,28 +1184,37 @@ static inline bool intel_pt_skip_event(struct intel_pt *pt)
 	       pt->num_events++ < pt->synth_opts.initial_skip;
 }
 
+static void intel_pt_prep_a_sample(struct intel_pt_queue *ptq,
+				   union perf_event *event,
+				   struct perf_sample *sample)
+{
+	event->sample.header.type = PERF_RECORD_SAMPLE;
+	event->sample.header.size = sizeof(struct perf_event_header);
+
+	sample->pid = ptq->pid;
+	sample->tid = ptq->tid;
+	sample->cpu = ptq->cpu;
+	sample->insn_len = ptq->insn_len;
+	memcpy(sample->insn, ptq->insn, INTEL_PT_INSN_BUF_SZ);
+}
+
 static void intel_pt_prep_b_sample(struct intel_pt *pt,
 				   struct intel_pt_queue *ptq,
 				   union perf_event *event,
 				   struct perf_sample *sample)
 {
+	intel_pt_prep_a_sample(ptq, event, sample);
+
 	if (!pt->timeless_decoding)
 		sample->time = tsc_to_perf_time(ptq->timestamp, &pt->tc);
 
 	sample->ip = ptq->state->from_ip;
 	sample->cpumode = intel_pt_cpumode(pt, sample->ip);
-	sample->pid = ptq->pid;
-	sample->tid = ptq->tid;
 	sample->addr = ptq->state->to_ip;
 	sample->period = 1;
-	sample->cpu = ptq->cpu;
 	sample->flags = ptq->flags;
-	sample->insn_len = ptq->insn_len;
-	memcpy(sample->insn, ptq->insn, INTEL_PT_INSN_BUF_SZ);
 
-	event->sample.header.type = PERF_RECORD_SAMPLE;
 	event->sample.header.misc = sample->cpumode;
-	event->sample.header.size = sizeof(struct perf_event_header);
 }
 
 static int intel_pt_inject_event(union perf_event *event,
@@ -1534,6 +1549,261 @@ static int intel_pt_synth_pwrx_sample(struct intel_pt_queue *ptq)
 					    pt->pwr_events_sample_type);
 }
 
+/*
+ * PEBS gp_regs array indexes plus 1 so that 0 means not present. Refer
+ * intel_pt_add_gp_regs().
+ */
+static const int pebs_gp_regs[] = {
+	[PERF_REG_X86_FLAGS]	= 1,
+	[PERF_REG_X86_IP]	= 2,
+	[PERF_REG_X86_AX]	= 3,
+	[PERF_REG_X86_CX]	= 4,
+	[PERF_REG_X86_DX]	= 5,
+	[PERF_REG_X86_BX]	= 6,
+	[PERF_REG_X86_SP]	= 7,
+	[PERF_REG_X86_BP]	= 8,
+	[PERF_REG_X86_SI]	= 9,
+	[PERF_REG_X86_DI]	= 10,
+	[PERF_REG_X86_R8]	= 11,
+	[PERF_REG_X86_R9]	= 12,
+	[PERF_REG_X86_R10]	= 13,
+	[PERF_REG_X86_R11]	= 14,
+	[PERF_REG_X86_R12]	= 15,
+	[PERF_REG_X86_R13]	= 16,
+	[PERF_REG_X86_R14]	= 17,
+	[PERF_REG_X86_R15]	= 18,
+};
+
+static u64 *intel_pt_add_gp_regs(struct regs_dump *intr_regs, u64 *pos,
+				 const struct intel_pt_blk_items *items,
+				 u64 regs_mask)
+{
+	const u64 *gp_regs = items->val[INTEL_PT_GP_REGS_POS];
+	u32 mask = items->mask[INTEL_PT_GP_REGS_POS];
+	u32 bit;
+	int i;
+
+	for (i = 0, bit = 1; i < PERF_REG_X86_64_MAX; i++, bit <<= 1) {
+		/* Get the PEBS gp_regs array index */
+		int n = pebs_gp_regs[i] - 1;
+
+		if (n < 0)
+			continue;
+		/*
+		 * Add only registers that were requested (i.e. 'regs_mask') and
+		 * that were provided (i.e. 'mask'), and update the resulting
+		 * mask (i.e. 'intr_regs->mask') accordingly.
+		 */
+		if (mask & 1 << n && regs_mask & bit) {
+			intr_regs->mask |= bit;
+			*pos++ = gp_regs[n];
+		}
+	}
+
+	return pos;
+}
+
+#ifndef PERF_REG_X86_XMM0
+#define PERF_REG_X86_XMM0 32
+#endif
+
+static void intel_pt_add_xmm(struct regs_dump *intr_regs, u64 *pos,
+			     const struct intel_pt_blk_items *items,
+			     u64 regs_mask)
+{
+	u32 mask = items->has_xmm & (regs_mask >> PERF_REG_X86_XMM0);
+	const u64 *xmm = items->xmm;
+
+	/*
+	 * If there are any XMM registers, then there should be all of them.
+	 * Nevertheless, follow the logic to add only registers that were
+	 * requested (i.e. 'regs_mask') and that were provided (i.e. 'mask'),
+	 * and update the resulting mask (i.e. 'intr_regs->mask') accordingly.
+	 */
+	intr_regs->mask |= (u64)mask << PERF_REG_X86_XMM0;
+
+	for (; mask; mask >>= 1, xmm++) {
+		if (mask & 1)
+			*pos++ = *xmm;
+	}
+}
+
+#define LBR_INFO_MISPRED	(1ULL << 63)
+#define LBR_INFO_IN_TX		(1ULL << 62)
+#define LBR_INFO_ABORT		(1ULL << 61)
+#define LBR_INFO_CYCLES		0xffff
+
+/* Refer kernel's intel_pmu_store_pebs_lbrs() */
+static u64 intel_pt_lbr_flags(u64 info)
+{
+	union {
+		struct branch_flags flags;
+		u64 result;
+	} u = {
+		.flags = {
+			.mispred	= !!(info & LBR_INFO_MISPRED),
+			.predicted	= !(info & LBR_INFO_MISPRED),
+			.in_tx		= !!(info & LBR_INFO_IN_TX),
+			.abort		= !!(info & LBR_INFO_ABORT),
+			.cycles		= info & LBR_INFO_CYCLES,
+		}
+	};
+
+	return u.result;
+}
+
+static void intel_pt_add_lbrs(struct branch_stack *br_stack,
+			      const struct intel_pt_blk_items *items)
+{
+	u64 *to;
+	int i;
+
+	br_stack->nr = 0;
+
+	to = &br_stack->entries[0].from;
+
+	for (i = INTEL_PT_LBR_0_POS; i <= INTEL_PT_LBR_2_POS; i++) {
+		u32 mask = items->mask[i];
+		const u64 *from = items->val[i];
+
+		for (; mask; mask >>= 3, from += 3) {
+			if ((mask & 7) == 7) {
+				*to++ = from[0];
+				*to++ = from[1];
+				*to++ = intel_pt_lbr_flags(from[2]);
+				br_stack->nr += 1;
+			}
+		}
+	}
+}
+
+/* INTEL_PT_LBR_0, INTEL_PT_LBR_1 and INTEL_PT_LBR_2 */
+#define LBRS_MAX (INTEL_PT_BLK_ITEM_ID_CNT * 3)
+
+static int intel_pt_synth_pebs_sample(struct intel_pt_queue *ptq)
+{
+	const struct intel_pt_blk_items *items = &ptq->state->items;
+	struct perf_sample sample = { .ip = 0, };
+	union perf_event *event = ptq->event_buf;
+	struct intel_pt *pt = ptq->pt;
+	struct perf_evsel *evsel = pt->pebs_evsel;
+	u64 sample_type = evsel->attr.sample_type;
+	u64 id = evsel->id[0];
+	u8 cpumode;
+
+	if (intel_pt_skip_event(pt))
+		return 0;
+
+	intel_pt_prep_a_sample(ptq, event, &sample);
+
+	sample.id = id;
+	sample.stream_id = id;
+
+	if (!evsel->attr.freq)
+		sample.period = evsel->attr.sample_period;
+
+	/* No support for non-zero CS base */
+	if (items->has_ip)
+		sample.ip = items->ip;
+	else if (items->has_rip)
+		sample.ip = items->rip;
+	else
+		sample.ip = ptq->state->from_ip;
+
+	/* No support for guest mode at this time */
+	cpumode = sample.ip < ptq->pt->kernel_start ?
+		  PERF_RECORD_MISC_USER :
+		  PERF_RECORD_MISC_KERNEL;
+
+	event->sample.header.misc = cpumode | PERF_RECORD_MISC_EXACT_IP;
+
+	sample.cpumode = cpumode;
+
+	if (sample_type & PERF_SAMPLE_TIME) {
+		u64 timestamp = 0;
+
+		if (items->has_timestamp)
+			timestamp = items->timestamp;
+		else if (!pt->timeless_decoding)
+			timestamp = ptq->timestamp;
+		if (timestamp)
+			sample.time = tsc_to_perf_time(timestamp, &pt->tc);
+	}
+
+	if (sample_type & PERF_SAMPLE_CALLCHAIN &&
+	    pt->synth_opts.callchain) {
+		thread_stack__sample(ptq->thread, ptq->cpu, ptq->chain,
+				     pt->synth_opts.callchain_sz, sample.ip,
+				     pt->kernel_start);
+		sample.callchain = ptq->chain;
+	}
+
+	if (sample_type & PERF_SAMPLE_REGS_INTR &&
+	    items->mask[INTEL_PT_GP_REGS_POS]) {
+		u64 regs[sizeof(sample.intr_regs.mask)];
+		u64 regs_mask = evsel->attr.sample_regs_intr;
+		u64 *pos;
+
+		sample.intr_regs.abi = items->is_32_bit ?
+				       PERF_SAMPLE_REGS_ABI_32 :
+				       PERF_SAMPLE_REGS_ABI_64;
+		sample.intr_regs.regs = regs;
+
+		pos = intel_pt_add_gp_regs(&sample.intr_regs, regs, items, regs_mask);
+
+		intel_pt_add_xmm(&sample.intr_regs, pos, items, regs_mask);
+	}
+
+	if (sample_type & PERF_SAMPLE_BRANCH_STACK) {
+		struct {
+			struct branch_stack br_stack;
+			struct branch_entry entries[LBRS_MAX];
+		} br;
+
+		if (items->mask[INTEL_PT_LBR_0_POS] ||
+		    items->mask[INTEL_PT_LBR_1_POS] ||
+		    items->mask[INTEL_PT_LBR_2_POS]) {
+			intel_pt_add_lbrs(&br.br_stack, items);
+			sample.branch_stack = &br.br_stack;
+		} else if (pt->synth_opts.last_branch) {
+			intel_pt_copy_last_branch_rb(ptq);
+			sample.branch_stack = ptq->last_branch;
+		} else {
+			br.br_stack.nr = 0;
+			sample.branch_stack = &br.br_stack;
+		}
+	}
+
+	if (sample_type & PERF_SAMPLE_ADDR && items->has_mem_access_address)
+		sample.addr = items->mem_access_address;
+
+	if (sample_type & PERF_SAMPLE_WEIGHT) {
+		/*
+		 * Refer kernel's setup_pebs_adaptive_sample_data() and
+		 * intel_hsw_weight().
+		 */
+		if (items->has_mem_access_latency)
+			sample.weight = items->mem_access_latency;
+		if (!sample.weight && items->has_tsx_aux_info) {
+			/* Cycles last block */
+			sample.weight = (u32)items->tsx_aux_info;
+		}
+	}
+
+	if (sample_type & PERF_SAMPLE_TRANSACTION && items->has_tsx_aux_info) {
+		u64 ax = items->has_rax ? items->rax : 0;
+		/* Refer kernel's intel_hsw_transaction() */
+		u64 txn = (u8)(items->tsx_aux_info >> 32);
+
+		/* For RTM XABORTs also log the abort code from AX */
+		if (txn & PERF_TXN_TRANSACTION && ax & 1)
+			txn |= ((ax >> 24) & 0xff) << PERF_TXN_ABORT_SHIFT;
+		sample.transaction = txn;
+	}
+
+	return intel_pt_deliver_synth_event(pt, ptq, event, &sample, sample_type);
+}
+
 static int intel_pt_synth_error(struct intel_pt *pt, int code, int cpu,
 				pid_t pid, pid_t tid, u64 ip, u64 timestamp)
 {
@@ -1621,6 +1891,16 @@ static int intel_pt_sample(struct intel_pt_queue *ptq)
 		ptq->ipc_cyc_cnt = ptq->state->tot_cyc_cnt;
 	}
 
+	/*
+	 * Do PEBS first to allow for the possibility that the PEBS timestamp
+	 * precedes the current timestamp.
+	 */
+	if (pt->sample_pebs && state->type & INTEL_PT_BLK_ITEMS) {
+		err = intel_pt_synth_pebs_sample(ptq);
+		if (err)
+			return err;
+	}
+
 	if (pt->sample_pwr_events && (state->type & INTEL_PT_PWR_EVT)) {
 		if (state->type & INTEL_PT_CBR_CHG) {
 			err = intel_pt_synth_cbr_sample(ptq);