perf tools: Add hw_idx in struct branch_stack
authorKan Liang <kan.liang@linux.intel.com>
Fri, 28 Feb 2020 16:30:00 +0000 (08:30 -0800)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Tue, 10 Mar 2020 00:42:53 +0000 (21:42 -0300)
The low level index of raw branch records for the most recent branch can
be recorded in a sample with PERF_SAMPLE_BRANCH_HW_INDEX
branch_sample_type. Extend struct branch_stack to support it.

However, if the PERF_SAMPLE_BRANCH_HW_INDEX is not applied, only nr and
entries[] will be output by kernel. The pointer of entries[] could be
wrong, since the output format is different with new struct
branch_stack.  Add a variable no_hw_idx in struct perf_sample to
indicate whether the hw_idx is output.  Add get_branch_entry() to return
corresponding pointer of entries[0].

To make dummy branch sample consistent as new branch sample, add hw_idx
in struct dummy_branch_stack for cs-etm and intel-pt.

Apply the new struct branch_stack for synthetic events as well.

Extend test case sample-parsing to support new struct branch_stack.

Committer notes:

Renamed get_branch_entries() to perf_sample__branch_entries() to have
proper namespacing and pave the way for this to be moved to libperf,
eventually.

Add 'static' to that inline as it is in a header.

Add 'hw_idx' to 'struct dummy_branch_stack' in cs-etm.c to fix the build
on arm64.

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexey Budankov <alexey.budankov@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Pavel Gerasimov <pavel.gerasimov@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Vitaly Slobodskoy <vitaly.slobodskoy@intel.com>
Link: http://lore.kernel.org/lkml/20200228163011.19358-2-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
13 files changed:
tools/perf/builtin-script.c
tools/perf/tests/sample-parsing.c
tools/perf/util/branch.h
tools/perf/util/cs-etm.c
tools/perf/util/event.h
tools/perf/util/evsel.c
tools/perf/util/evsel.h
tools/perf/util/hist.c
tools/perf/util/intel-pt.c
tools/perf/util/machine.c
tools/perf/util/scripting-engines/trace-event-python.c
tools/perf/util/session.c
tools/perf/util/synthetic-events.c

index e2406b291c1c57a8a880d6249776909c7c4781b2..656b347f6dd84876a2f82ce5985265ba2d45597e 100644 (file)
@@ -735,6 +735,7 @@ static int perf_sample__fprintf_brstack(struct perf_sample *sample,
                                        struct perf_event_attr *attr, FILE *fp)
 {
        struct branch_stack *br = sample->branch_stack;
+       struct branch_entry *entries = perf_sample__branch_entries(sample);
        struct addr_location alf, alt;
        u64 i, from, to;
        int printed = 0;
@@ -743,8 +744,8 @@ static int perf_sample__fprintf_brstack(struct perf_sample *sample,
                return 0;
 
        for (i = 0; i < br->nr; i++) {
-               from = br->entries[i].from;
-               to   = br->entries[i].to;
+               from = entries[i].from;
+               to   = entries[i].to;
 
                if (PRINT_FIELD(DSO)) {
                        memset(&alf, 0, sizeof(alf));
@@ -768,10 +769,10 @@ static int perf_sample__fprintf_brstack(struct perf_sample *sample,
                }
 
                printed += fprintf(fp, "/%c/%c/%c/%d ",
-                       mispred_str( br->entries + i),
-                       br->entries[i].flags.in_tx? 'X' : '-',
-                       br->entries[i].flags.abort? 'A' : '-',
-                       br->entries[i].flags.cycles);
+                       mispred_str(entries + i),
+                       entries[i].flags.in_tx ? 'X' : '-',
+                       entries[i].flags.abort ? 'A' : '-',
+                       entries[i].flags.cycles);
        }
 
        return printed;
@@ -782,6 +783,7 @@ static int perf_sample__fprintf_brstacksym(struct perf_sample *sample,
                                           struct perf_event_attr *attr, FILE *fp)
 {
        struct branch_stack *br = sample->branch_stack;
+       struct branch_entry *entries = perf_sample__branch_entries(sample);
        struct addr_location alf, alt;
        u64 i, from, to;
        int printed = 0;
@@ -793,8 +795,8 @@ static int perf_sample__fprintf_brstacksym(struct perf_sample *sample,
 
                memset(&alf, 0, sizeof(alf));
                memset(&alt, 0, sizeof(alt));
-               from = br->entries[i].from;
-               to   = br->entries[i].to;
+               from = entries[i].from;
+               to   = entries[i].to;
 
                thread__find_symbol_fb(thread, sample->cpumode, from, &alf);
                thread__find_symbol_fb(thread, sample->cpumode, to, &alt);
@@ -813,10 +815,10 @@ static int perf_sample__fprintf_brstacksym(struct perf_sample *sample,
                        printed += fprintf(fp, ")");
                }
                printed += fprintf(fp, "/%c/%c/%c/%d ",
-                       mispred_str( br->entries + i),
-                       br->entries[i].flags.in_tx? 'X' : '-',
-                       br->entries[i].flags.abort? 'A' : '-',
-                       br->entries[i].flags.cycles);
+                       mispred_str(entries + i),
+                       entries[i].flags.in_tx ? 'X' : '-',
+                       entries[i].flags.abort ? 'A' : '-',
+                       entries[i].flags.cycles);
        }
 
        return printed;
@@ -827,6 +829,7 @@ static int perf_sample__fprintf_brstackoff(struct perf_sample *sample,
                                           struct perf_event_attr *attr, FILE *fp)
 {
        struct branch_stack *br = sample->branch_stack;
+       struct branch_entry *entries = perf_sample__branch_entries(sample);
        struct addr_location alf, alt;
        u64 i, from, to;
        int printed = 0;
@@ -838,8 +841,8 @@ static int perf_sample__fprintf_brstackoff(struct perf_sample *sample,
 
                memset(&alf, 0, sizeof(alf));
                memset(&alt, 0, sizeof(alt));
-               from = br->entries[i].from;
-               to   = br->entries[i].to;
+               from = entries[i].from;
+               to   = entries[i].to;
 
                if (thread__find_map_fb(thread, sample->cpumode, from, &alf) &&
                    !alf.map->dso->adjust_symbols)
@@ -862,10 +865,10 @@ static int perf_sample__fprintf_brstackoff(struct perf_sample *sample,
                        printed += fprintf(fp, ")");
                }
                printed += fprintf(fp, "/%c/%c/%c/%d ",
-                       mispred_str(br->entries + i),
-                       br->entries[i].flags.in_tx ? 'X' : '-',
-                       br->entries[i].flags.abort ? 'A' : '-',
-                       br->entries[i].flags.cycles);
+                       mispred_str(entries + i),
+                       entries[i].flags.in_tx ? 'X' : '-',
+                       entries[i].flags.abort ? 'A' : '-',
+                       entries[i].flags.cycles);
        }
 
        return printed;
@@ -1053,6 +1056,7 @@ static int perf_sample__fprintf_brstackinsn(struct perf_sample *sample,
                                            struct machine *machine, FILE *fp)
 {
        struct branch_stack *br = sample->branch_stack;
+       struct branch_entry *entries = perf_sample__branch_entries(sample);
        u64 start, end;
        int i, insn, len, nr, ilen, printed = 0;
        struct perf_insn x;
@@ -1073,31 +1077,31 @@ static int perf_sample__fprintf_brstackinsn(struct perf_sample *sample,
        printed += fprintf(fp, "%c", '\n');
 
        /* Handle first from jump, of which we don't know the entry. */
-       len = grab_bb(buffer, br->entries[nr-1].from,
-                       br->entries[nr-1].from,
+       len = grab_bb(buffer, entries[nr-1].from,
+                       entries[nr-1].from,
                        machine, thread, &x.is64bit, &x.cpumode, false);
        if (len > 0) {
-               printed += ip__fprintf_sym(br->entries[nr - 1].from, thread,
+               printed += ip__fprintf_sym(entries[nr - 1].from, thread,
                                           x.cpumode, x.cpu, &lastsym, attr, fp);
-               printed += ip__fprintf_jump(br->entries[nr - 1].from, &br->entries[nr - 1],
+               printed += ip__fprintf_jump(entries[nr - 1].from, &entries[nr - 1],
                                            &x, buffer, len, 0, fp, &total_cycles);
                if (PRINT_FIELD(SRCCODE))
-                       printed += print_srccode(thread, x.cpumode, br->entries[nr - 1].from);
+                       printed += print_srccode(thread, x.cpumode, entries[nr - 1].from);
        }
 
        /* Print all blocks */
        for (i = nr - 2; i >= 0; i--) {
-               if (br->entries[i].from || br->entries[i].to)
+               if (entries[i].from || entries[i].to)
                        pr_debug("%d: %" PRIx64 "-%" PRIx64 "\n", i,
-                                br->entries[i].from,
-                                br->entries[i].to);
-               start = br->entries[i + 1].to;
-               end   = br->entries[i].from;
+                                entries[i].from,
+                                entries[i].to);
+               start = entries[i + 1].to;
+               end   = entries[i].from;
 
                len = grab_bb(buffer, start, end, machine, thread, &x.is64bit, &x.cpumode, false);
                /* Patch up missing kernel transfers due to ring filters */
                if (len == -ENXIO && i > 0) {
-                       end = br->entries[--i].from;
+                       end = entries[--i].from;
                        pr_debug("\tpatching up to %" PRIx64 "-%" PRIx64 "\n", start, end);
                        len = grab_bb(buffer, start, end, machine, thread, &x.is64bit, &x.cpumode, false);
                }
@@ -1110,7 +1114,7 @@ static int perf_sample__fprintf_brstackinsn(struct perf_sample *sample,
 
                        printed += ip__fprintf_sym(ip, thread, x.cpumode, x.cpu, &lastsym, attr, fp);
                        if (ip == end) {
-                               printed += ip__fprintf_jump(ip, &br->entries[i], &x, buffer + off, len - off, ++insn, fp,
+                               printed += ip__fprintf_jump(ip, &entries[i], &x, buffer + off, len - off, ++insn, fp,
                                                            &total_cycles);
                                if (PRINT_FIELD(SRCCODE))
                                        printed += print_srccode(thread, x.cpumode, ip);
@@ -1134,9 +1138,9 @@ static int perf_sample__fprintf_brstackinsn(struct perf_sample *sample,
         * Hit the branch? In this case we are already done, and the target
         * has not been executed yet.
         */
-       if (br->entries[0].from == sample->ip)
+       if (entries[0].from == sample->ip)
                goto out;
-       if (br->entries[0].flags.abort)
+       if (entries[0].flags.abort)
                goto out;
 
        /*
@@ -1147,7 +1151,7 @@ static int perf_sample__fprintf_brstackinsn(struct perf_sample *sample,
         * between final branch and sample. When this happens just
         * continue walking after the last TO until we hit a branch.
         */
-       start = br->entries[0].to;
+       start = entries[0].to;
        end = sample->ip;
        if (end < start) {
                /* Missing jump. Scan 128 bytes for the next branch */
index 2762e11552380e7350e1de4dd9b0413db0549d1b..14239e4721870804096c589967a7c3046fab6138 100644 (file)
@@ -99,6 +99,7 @@ static bool samples_same(const struct perf_sample *s1,
 
        if (type & PERF_SAMPLE_BRANCH_STACK) {
                COMP(branch_stack->nr);
+               COMP(branch_stack->hw_idx);
                for (i = 0; i < s1->branch_stack->nr; i++)
                        MCOMP(branch_stack->entries[i]);
        }
@@ -186,7 +187,7 @@ static int do_test(u64 sample_type, u64 sample_regs, u64 read_format)
                u64 data[64];
        } branch_stack = {
                /* 1 branch_entry */
-               .data = {1, 211, 212, 213},
+               .data = {1, -1ULL, 211, 212, 213},
        };
        u64 regs[64];
        const u64 raw_data[] = {0x123456780a0b0c0dULL, 0x1102030405060708ULL};
@@ -208,6 +209,7 @@ static int do_test(u64 sample_type, u64 sample_regs, u64 read_format)
                .transaction    = 112,
                .raw_data       = (void *)raw_data,
                .callchain      = &callchain.callchain,
+               .no_hw_idx      = false,
                .branch_stack   = &branch_stack.branch_stack,
                .user_regs      = {
                        .abi    = PERF_SAMPLE_REGS_ABI_64,
@@ -244,6 +246,9 @@ static int do_test(u64 sample_type, u64 sample_regs, u64 read_format)
        if (sample_type & PERF_SAMPLE_REGS_INTR)
                evsel.core.attr.sample_regs_intr = sample_regs;
 
+       if (sample_type & PERF_SAMPLE_BRANCH_STACK)
+               evsel.core.attr.branch_sample_type |= PERF_SAMPLE_BRANCH_HW_INDEX;
+
        for (i = 0; i < sizeof(regs); i++)
                *(i + (u8 *)regs) = i & 0xfe;
 
index 88e00d268f6f2795a0f0b311a8d0d2de1fd59c56..154a05cd03af5c66023f6cfb78e90ccdd226c85b 100644 (file)
@@ -12,6 +12,7 @@
 #include <linux/stddef.h>
 #include <linux/perf_event.h>
 #include <linux/types.h>
+#include "event.h"
 
 struct branch_flags {
        u64 mispred:1;
@@ -39,9 +40,30 @@ struct branch_entry {
 
 struct branch_stack {
        u64                     nr;
+       u64                     hw_idx;
        struct branch_entry     entries[0];
 };
 
+/*
+ * The hw_idx is only available when PERF_SAMPLE_BRANCH_HW_INDEX is applied.
+ * Otherwise, the output format of a sample with branch stack is
+ * struct branch_stack {
+ *     u64                     nr;
+ *     struct branch_entry     entries[0];
+ * }
+ * Check whether the hw_idx is available,
+ * and return the corresponding pointer of entries[0].
+ */
+static inline struct branch_entry *perf_sample__branch_entries(struct perf_sample *sample)
+{
+       u64 *entry = (u64 *)sample->branch_stack;
+
+       entry++;
+       if (sample->no_hw_idx)
+               return (struct branch_entry *)entry;
+       return (struct branch_entry *)(++entry);
+}
+
 struct branch_type_stat {
        bool    branch_to;
        u64     counts[PERF_BR_MAX];
index 5471045ebf5c88e3760759739b1aa7069c95b95a..b3b3fe3ea3457da62a6e243799ba40a1c15bf825 100644 (file)
@@ -1172,6 +1172,7 @@ static int cs_etm__synth_branch_sample(struct cs_etm_queue *etmq,
        union perf_event *event = tidq->event_buf;
        struct dummy_branch_stack {
                u64                     nr;
+               u64                     hw_idx;
                struct branch_entry     entries;
        } dummy_bs;
        u64 ip;
@@ -1202,6 +1203,7 @@ static int cs_etm__synth_branch_sample(struct cs_etm_queue *etmq,
        if (etm->synth_opts.last_branch) {
                dummy_bs = (struct dummy_branch_stack){
                        .nr = 1,
+                       .hw_idx = -1ULL,
                        .entries = {
                                .from = sample.ip,
                                .to = sample.addr,
index 85223159737c1be1554dd6ca8c8d95495d304722..3cda40a2fafc9b90fc0af61ee9dbefb744c0e4de 100644 (file)
@@ -139,6 +139,7 @@ struct perf_sample {
        u16 insn_len;
        u8  cpumode;
        u16 misc;
+       bool no_hw_idx;         /* No hw_idx collected in branch_stack */
        char insn[MAX_INSN];
        void *raw_data;
        struct ip_callchain *callchain;
index c8dc4450884cd9ddbbfcb8578ded56982ad7ab16..05883a45de5b30f2a978dc75be627e4b36b3bd76 100644 (file)
@@ -2169,7 +2169,12 @@ int perf_evsel__parse_sample(struct evsel *evsel, union perf_event *event,
 
                if (data->branch_stack->nr > max_branch_nr)
                        return -EFAULT;
+
                sz = data->branch_stack->nr * sizeof(struct branch_entry);
+               if (perf_evsel__has_branch_hw_idx(evsel))
+                       sz += sizeof(u64);
+               else
+                       data->no_hw_idx = true;
                OVERFLOW_CHECK(array, sz, max_size);
                array = (void *)array + sz;
        }
index dc14f4a823cd6a1cee5e2f9edae77cb2ac2e92bb..99a0cb60c556361d4e7c466aa9df20c0bf30fd37 100644 (file)
@@ -389,6 +389,11 @@ static inline bool perf_evsel__has_branch_callstack(const struct evsel *evsel)
        return evsel->core.attr.branch_sample_type & PERF_SAMPLE_BRANCH_CALL_STACK;
 }
 
+static inline bool perf_evsel__has_branch_hw_idx(const struct evsel *evsel)
+{
+       return evsel->core.attr.branch_sample_type & PERF_SAMPLE_BRANCH_HW_INDEX;
+}
+
 static inline bool evsel__has_callchain(const struct evsel *evsel)
 {
        return (evsel->core.attr.sample_type & PERF_SAMPLE_CALLCHAIN) != 0;
index ca5a8f4d007ec7d689948cf739933ea24555cde4..e74a5acf66d948ebd424857686c081a43da5ae58 100644 (file)
@@ -2584,9 +2584,10 @@ void hist__account_cycles(struct branch_stack *bs, struct addr_location *al,
                          u64 *total_cycles)
 {
        struct branch_info *bi;
+       struct branch_entry *entries = perf_sample__branch_entries(sample);
 
        /* If we have branch cycles always annotate them. */
-       if (bs && bs->nr && bs->entries[0].flags.cycles) {
+       if (bs && bs->nr && entries[0].flags.cycles) {
                int i;
 
                bi = sample__resolve_bstack(sample, al);
index 33cf8928cf059c04a32147662d5258eda2079213..23c8289c2472d723e7a07413f3705c3c60c3cb94 100644 (file)
@@ -1295,6 +1295,7 @@ static int intel_pt_synth_branch_sample(struct intel_pt_queue *ptq)
        struct perf_sample sample = { .ip = 0, };
        struct dummy_branch_stack {
                u64                     nr;
+               u64                     hw_idx;
                struct branch_entry     entries;
        } dummy_bs;
 
@@ -1316,6 +1317,7 @@ static int intel_pt_synth_branch_sample(struct intel_pt_queue *ptq)
        if (pt->synth_opts.last_branch && sort__mode == SORT_MODE__BRANCH) {
                dummy_bs = (struct dummy_branch_stack){
                        .nr = 1,
+                       .hw_idx = -1ULL,
                        .entries = {
                                .from = sample.ip,
                                .to = sample.addr,
index fb5c2cd44d3003ac1c0dc1621d8cbb718ab1bdd0..fd14f1489802040b8b7f30e0724608ac3891dff5 100644 (file)
@@ -2081,15 +2081,16 @@ struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
 {
        unsigned int i;
        const struct branch_stack *bs = sample->branch_stack;
+       struct branch_entry *entries = perf_sample__branch_entries(sample);
        struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
 
        if (!bi)
                return NULL;
 
        for (i = 0; i < bs->nr; i++) {
-               ip__resolve_ams(al->thread, &bi[i].to, bs->entries[i].to);
-               ip__resolve_ams(al->thread, &bi[i].from, bs->entries[i].from);
-               bi[i].flags = bs->entries[i].flags;
+               ip__resolve_ams(al->thread, &bi[i].to, entries[i].to);
+               ip__resolve_ams(al->thread, &bi[i].from, entries[i].from);
+               bi[i].flags = entries[i].flags;
        }
        return bi;
 }
@@ -2185,6 +2186,7 @@ static int resolve_lbr_callchain_sample(struct thread *thread,
        /* LBR only affects the user callchain */
        if (i != chain_nr) {
                struct branch_stack *lbr_stack = sample->branch_stack;
+               struct branch_entry *entries = perf_sample__branch_entries(sample);
                int lbr_nr = lbr_stack->nr, j, k;
                bool branch;
                struct branch_flags *flags;
@@ -2210,31 +2212,29 @@ static int resolve_lbr_callchain_sample(struct thread *thread,
                                        ip = chain->ips[j];
                                else if (j > i + 1) {
                                        k = j - i - 2;
-                                       ip = lbr_stack->entries[k].from;
+                                       ip = entries[k].from;
                                        branch = true;
-                                       flags = &lbr_stack->entries[k].flags;
+                                       flags = &entries[k].flags;
                                } else {
-                                       ip = lbr_stack->entries[0].to;
+                                       ip = entries[0].to;
                                        branch = true;
-                                       flags = &lbr_stack->entries[0].flags;
-                                       branch_from =
-                                               lbr_stack->entries[0].from;
+                                       flags = &entries[0].flags;
+                                       branch_from = entries[0].from;
                                }
                        } else {
                                if (j < lbr_nr) {
                                        k = lbr_nr - j - 1;
-                                       ip = lbr_stack->entries[k].from;
+                                       ip = entries[k].from;
                                        branch = true;
-                                       flags = &lbr_stack->entries[k].flags;
+                                       flags = &entries[k].flags;
                                }
                                else if (j > lbr_nr)
                                        ip = chain->ips[i + 1 - (j - lbr_nr)];
                                else {
-                                       ip = lbr_stack->entries[0].to;
+                                       ip = entries[0].to;
                                        branch = true;
-                                       flags = &lbr_stack->entries[0].flags;
-                                       branch_from =
-                                               lbr_stack->entries[0].from;
+                                       flags = &entries[0].flags;
+                                       branch_from = entries[0].from;
                                }
                        }
 
@@ -2281,6 +2281,7 @@ static int thread__resolve_callchain_sample(struct thread *thread,
                                            int max_stack)
 {
        struct branch_stack *branch = sample->branch_stack;
+       struct branch_entry *entries = perf_sample__branch_entries(sample);
        struct ip_callchain *chain = sample->callchain;
        int chain_nr = 0;
        u8 cpumode = PERF_RECORD_MISC_USER;
@@ -2328,7 +2329,7 @@ static int thread__resolve_callchain_sample(struct thread *thread,
 
                for (i = 0; i < nr; i++) {
                        if (callchain_param.order == ORDER_CALLEE) {
-                               be[i] = branch->entries[i];
+                               be[i] = entries[i];
 
                                if (chain == NULL)
                                        continue;
@@ -2347,7 +2348,7 @@ static int thread__resolve_callchain_sample(struct thread *thread,
                                    be[i].from >= chain->ips[first_call] - 8)
                                        first_call++;
                        } else
-                               be[i] = branch->entries[branch->nr - i - 1];
+                               be[i] = entries[branch->nr - i - 1];
                }
 
                memset(iter, 0, sizeof(struct iterations) * nr);
index 80ca5d0ab7fe21c9d509e66a5f0a33299aa6cfd6..8c1b27cd8b99235035f8a9e3b4b50fdecefe524a 100644 (file)
@@ -464,6 +464,7 @@ static PyObject *python_process_brstack(struct perf_sample *sample,
                                        struct thread *thread)
 {
        struct branch_stack *br = sample->branch_stack;
+       struct branch_entry *entries = perf_sample__branch_entries(sample);
        PyObject *pylist;
        u64 i;
 
@@ -484,28 +485,28 @@ static PyObject *python_process_brstack(struct perf_sample *sample,
                        Py_FatalError("couldn't create Python dictionary");
 
                pydict_set_item_string_decref(pyelem, "from",
-                   PyLong_FromUnsignedLongLong(br->entries[i].from));
+                   PyLong_FromUnsignedLongLong(entries[i].from));
                pydict_set_item_string_decref(pyelem, "to",
-                   PyLong_FromUnsignedLongLong(br->entries[i].to));
+                   PyLong_FromUnsignedLongLong(entries[i].to));
                pydict_set_item_string_decref(pyelem, "mispred",
-                   PyBool_FromLong(br->entries[i].flags.mispred));
+                   PyBool_FromLong(entries[i].flags.mispred));
                pydict_set_item_string_decref(pyelem, "predicted",
-                   PyBool_FromLong(br->entries[i].flags.predicted));
+                   PyBool_FromLong(entries[i].flags.predicted));
                pydict_set_item_string_decref(pyelem, "in_tx",
-                   PyBool_FromLong(br->entries[i].flags.in_tx));
+                   PyBool_FromLong(entries[i].flags.in_tx));
                pydict_set_item_string_decref(pyelem, "abort",
-                   PyBool_FromLong(br->entries[i].flags.abort));
+                   PyBool_FromLong(entries[i].flags.abort));
                pydict_set_item_string_decref(pyelem, "cycles",
-                   PyLong_FromUnsignedLongLong(br->entries[i].flags.cycles));
+                   PyLong_FromUnsignedLongLong(entries[i].flags.cycles));
 
                thread__find_map_fb(thread, sample->cpumode,
-                                   br->entries[i].from, &al);
+                                   entries[i].from, &al);
                dsoname = get_dsoname(al.map);
                pydict_set_item_string_decref(pyelem, "from_dsoname",
                                              _PyUnicode_FromString(dsoname));
 
                thread__find_map_fb(thread, sample->cpumode,
-                                   br->entries[i].to, &al);
+                                   entries[i].to, &al);
                dsoname = get_dsoname(al.map);
                pydict_set_item_string_decref(pyelem, "to_dsoname",
                                              _PyUnicode_FromString(dsoname));
@@ -561,6 +562,7 @@ static PyObject *python_process_brstacksym(struct perf_sample *sample,
                                           struct thread *thread)
 {
        struct branch_stack *br = sample->branch_stack;
+       struct branch_entry *entries = perf_sample__branch_entries(sample);
        PyObject *pylist;
        u64 i;
        char bf[512];
@@ -581,22 +583,22 @@ static PyObject *python_process_brstacksym(struct perf_sample *sample,
                        Py_FatalError("couldn't create Python dictionary");
 
                thread__find_symbol_fb(thread, sample->cpumode,
-                                      br->entries[i].from, &al);
+                                      entries[i].from, &al);
                get_symoff(al.sym, &al, true, bf, sizeof(bf));
                pydict_set_item_string_decref(pyelem, "from",
                                              _PyUnicode_FromString(bf));
 
                thread__find_symbol_fb(thread, sample->cpumode,
-                                      br->entries[i].to, &al);
+                                      entries[i].to, &al);
                get_symoff(al.sym, &al, true, bf, sizeof(bf));
                pydict_set_item_string_decref(pyelem, "to",
                                              _PyUnicode_FromString(bf));
 
-               get_br_mspred(&br->entries[i].flags, bf, sizeof(bf));
+               get_br_mspred(&entries[i].flags, bf, sizeof(bf));
                pydict_set_item_string_decref(pyelem, "pred",
                                              _PyUnicode_FromString(bf));
 
-               if (br->entries[i].flags.in_tx) {
+               if (entries[i].flags.in_tx) {
                        pydict_set_item_string_decref(pyelem, "in_tx",
                                              _PyUnicode_FromString("X"));
                } else {
@@ -604,7 +606,7 @@ static PyObject *python_process_brstacksym(struct perf_sample *sample,
                                              _PyUnicode_FromString("-"));
                }
 
-               if (br->entries[i].flags.abort) {
+               if (entries[i].flags.abort) {
                        pydict_set_item_string_decref(pyelem, "abort",
                                              _PyUnicode_FromString("A"));
                } else {
index d0d7d25b23e3b049fc708132d7ff989117b71b9d..055b00abd56d855b343f852d2139f41d360d0f6a 100644 (file)
@@ -1007,6 +1007,7 @@ static void callchain__lbr_callstack_printf(struct perf_sample *sample)
 {
        struct ip_callchain *callchain = sample->callchain;
        struct branch_stack *lbr_stack = sample->branch_stack;
+       struct branch_entry *entries = perf_sample__branch_entries(sample);
        u64 kernel_callchain_nr = callchain->nr;
        unsigned int i;
 
@@ -1043,10 +1044,10 @@ static void callchain__lbr_callstack_printf(struct perf_sample *sample)
                               i, callchain->ips[i]);
 
                printf("..... %2d: %016" PRIx64 "\n",
-                      (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
+                      (int)(kernel_callchain_nr), entries[0].to);
                for (i = 0; i < lbr_stack->nr; i++)
                        printf("..... %2d: %016" PRIx64 "\n",
-                              (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
+                              (int)(i + kernel_callchain_nr + 1), entries[i].from);
        }
 }
 
@@ -1068,6 +1069,7 @@ static void callchain__printf(struct evsel *evsel,
 
 static void branch_stack__printf(struct perf_sample *sample, bool callstack)
 {
+       struct branch_entry *entries = perf_sample__branch_entries(sample);
        uint64_t i;
 
        printf("%s: nr:%" PRIu64 "\n",
@@ -1075,7 +1077,7 @@ static void branch_stack__printf(struct perf_sample *sample, bool callstack)
                sample->branch_stack->nr);
 
        for (i = 0; i < sample->branch_stack->nr; i++) {
-               struct branch_entry *e = &sample->branch_stack->entries[i];
+               struct branch_entry *e = &entries[i];
 
                if (!callstack) {
                        printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
index c423298fe62db81e26e30cf488a02b4dffa24d23..dd3e6f43fb86284eb80faea3d43c1ac3272bea6d 100644 (file)
@@ -1183,7 +1183,8 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
 
        if (type & PERF_SAMPLE_BRANCH_STACK) {
                sz = sample->branch_stack->nr * sizeof(struct branch_entry);
-               sz += sizeof(u64);
+               /* nr, hw_idx */
+               sz += 2 * sizeof(u64);
                result += sz;
        }
 
@@ -1344,7 +1345,8 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type, u64 read_fo
 
        if (type & PERF_SAMPLE_BRANCH_STACK) {
                sz = sample->branch_stack->nr * sizeof(struct branch_entry);
-               sz += sizeof(u64);
+               /* nr, hw_idx */
+               sz += 2 * sizeof(u64);
                memcpy(array, sample->branch_stack, sz);
                array = (void *)array + sz;
        }