aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2014-03-31 14:13:25 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2014-03-31 14:13:25 -0400
commit8c292f11744297dfb3a69f4a0bccbe4a6417b50d (patch)
treef1a89560de25a69b697d459a9b5cf2e738038d9f
parentd31605dc8a63f1df28443ddb3560b1079417af92 (diff)
parent538592ff0b008237ae88f5ce5fb1247127dc3ce5 (diff)
Merge branch 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull perf changes from Ingo Molnar: "Main changes: Kernel side changes: - Add SNB/IVB/HSW client uncore memory controller support (Stephane Eranian) - Fix various x86/P4 PMU driver bugs (Don Zickus) Tooling, user visible changes: - Add several futex 'perf bench' microbenchmarks (Davidlohr Bueso) - Speed up thread map generation (Don Zickus) - Introduce 'perf kvm --list-cmds' command line option for use by scripts (Ramkumar Ramachandra) - Print the evsel name in the annotate stdio output, prep to fix support outputting annotation for multiple events, not just for the first one (Arnaldo Carvalho de Melo) - Allow setting preferred callchain method in .perfconfig (Jiri Olsa) - Show in what binaries/modules 'perf probe's are set (Masami Hiramatsu) - Support distro-style debuginfo for uprobe in 'perf probe' (Masami Hiramatsu) Tooling, internal changes and fixes: - Use tid in mmap/mmap2 events to find maps (Don Zickus) - Record the reason for filtering an address_location (Namhyung Kim) - Apply all filters to an addr_location (Namhyung Kim) - Merge al->filtered with hist_entry->filtered in report/hists (Namhyung Kim) - Fix memory leak when synthesizing thread records (Namhyung Kim) - Use ui__has_annotation() in 'report' (Namhyung Kim) - hists browser refactorings to reuse code accross UIs (Namhyung Kim) - Add support for the new DWARF unwinder library in elfutils (Jiri Olsa) - Fix build race in the generation of bison files (Jiri Olsa) - Further streamline the feature detection display, trimming it a bit to show just the libraries detected, using VF=1 gets a more verbose output, showing the less interesting feature checks as well (Jiri Olsa). - Check compatible symtab type before loading dso (Namhyung Kim) - Check return value of filename__read_debuglink() (Stephane Eranian) - Move some hashing and fs related code from tools/perf/util/ to tools/lib/ so that it can be used by more tools/ living utilities (Borislav Petkov) - Prepare DWARF unwinding code for using an elfutils alternative unwinding library (Jiri Olsa) - Fix DWARF unwind max_stack processing (Jiri Olsa) - Add dwarf unwind 'perf test' entry (Jiri Olsa) - 'perf probe' improvements including memory leak fixes, sharing the intlist class with other tools, uprobes/kprobes code sharing and use of ref_reloc_sym (Masami Hiramatsu) - Shorten sample symbol resolving by adding cpumode to struct addr_location (Arnaldo Carvalho de Melo) - Fix synthesizing mmaps for threads (Don Zickus) - Fix invalid output on event group stdio report (Namhyung Kim) - Fixup header alignment in 'perf sched latency' output (Ramkumar Ramachandra) - Fix off-by-one error in 'perf timechart record' argv handling (Ramkumar Ramachandra) Tooling, cleanups: - Remove unused thread__find_map function (Jiri Olsa) - Remove unused simple_strtoul() function (Ramkumar Ramachandra) Tooling, documentation updates: - Update function names in debug messages (Ramkumar Ramachandra) - Update some code references in design.txt (Ramkumar Ramachandra) - Clarify load-latency information in the 'perf mem' docs (Andi Kleen) - Clarify x86 register naming in 'perf probe' docs (Andi Kleen)" * 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (96 commits) perf tools: Remove unused simple_strtoul() function perf tools: Update some code references in design.txt perf evsel: Update function names in debug messages perf tools: Remove thread__find_map function perf annotate: Print the evsel name in the stdio output perf report: Use ui__has_annotation() perf tools: Fix memory leak when synthesizing thread records perf tools: Use tid in mmap/mmap2 events to find maps perf report: Merge al->filtered with hist_entry->filtered perf symbols: Apply all filters to an addr_location perf symbols: Record the reason for filtering an address_location perf sched: Fixup header alignment in 'latency' output perf timechart: Fix off-by-one error in 'record' argv handling perf machine: Factor machine__find_thread to take tid argument perf tools: Speed up thread map generation perf kvm: introduce --list-cmds for use by scripts perf ui hists: Pass evsel to hpp->header/width functions explicitly perf symbols: Introduce thread__find_cpumode_addr_location perf session: Change header.misc dump from decimal to hex perf ui/tui: Reuse generic __hpp__fmt() code ...
-rw-r--r--arch/x86/include/asm/nmi.h3
-rw-r--r--arch/x86/kernel/cpu/perf_event.c47
-rw-r--r--arch/x86/kernel/cpu/perf_event.h8
-rw-r--r--arch/x86/kernel/cpu/perf_event_intel_uncore.c544
-rw-r--r--arch/x86/kernel/cpu/perf_event_intel_uncore.h5
-rw-r--r--arch/x86/kernel/cpu/perf_event_p4.c34
-rw-r--r--arch/x86/kernel/nmi.c37
-rw-r--r--include/linux/irq_work.h4
-rw-r--r--include/linux/pci_ids.h3
-rw-r--r--kernel/events/core.c39
-rw-r--r--kernel/irq_work.c6
-rw-r--r--kernel/trace/trace_event_perf.c22
-rw-r--r--tools/include/linux/hash.h5
-rw-r--r--tools/lib/api/Makefile2
-rw-r--r--tools/lib/api/fs/fs.c (renamed from tools/perf/util/fs.c)11
-rw-r--r--tools/lib/api/fs/fs.h (renamed from tools/perf/util/include/linux/magic.h)12
-rw-r--r--tools/perf/Documentation/perf-mem.txt4
-rw-r--r--tools/perf/Documentation/perf-probe.txt2
-rw-r--r--tools/perf/MANIFEST1
-rw-r--r--tools/perf/Makefile.perf42
-rw-r--r--tools/perf/arch/arm/Makefile2
-rw-r--r--tools/perf/arch/arm/util/unwind-libunwind.c (renamed from tools/perf/arch/arm/util/unwind.c)2
-rw-r--r--tools/perf/arch/x86/Makefile9
-rw-r--r--tools/perf/arch/x86/include/perf_regs.h6
-rw-r--r--tools/perf/arch/x86/tests/dwarf-unwind.c59
-rw-r--r--tools/perf/arch/x86/tests/regs_load.S92
-rw-r--r--tools/perf/arch/x86/util/unwind-libdw.c51
-rw-r--r--tools/perf/arch/x86/util/unwind-libunwind.c (renamed from tools/perf/arch/x86/util/unwind.c)4
-rw-r--r--tools/perf/bench/bench.h3
-rw-r--r--tools/perf/bench/futex-hash.c212
-rw-r--r--tools/perf/bench/futex-requeue.c211
-rw-r--r--tools/perf/bench/futex-wake.c201
-rw-r--r--tools/perf/bench/futex.h71
-rw-r--r--tools/perf/builtin-bench.c12
-rw-r--r--tools/perf/builtin-diff.c7
-rw-r--r--tools/perf/builtin-inject.c1
-rw-r--r--tools/perf/builtin-kvm.c12
-rw-r--r--tools/perf/builtin-probe.c12
-rw-r--r--tools/perf/builtin-record.c30
-rw-r--r--tools/perf/builtin-report.c28
-rw-r--r--tools/perf/builtin-sched.c10
-rw-r--r--tools/perf/builtin-timechart.c4
-rw-r--r--tools/perf/builtin-top.c12
-rw-r--r--tools/perf/config/Makefile232
-rw-r--r--tools/perf/config/feature-checks/Makefile6
-rw-r--r--tools/perf/config/feature-checks/test-all.c5
-rw-r--r--tools/perf/config/feature-checks/test-libdw-dwarf-unwind.c13
-rw-r--r--tools/perf/design.txt12
-rw-r--r--tools/perf/perf-completion.sh2
-rw-r--r--tools/perf/perf.h10
-rw-r--r--tools/perf/tests/builtin-test.c8
-rw-r--r--tools/perf/tests/dwarf-unwind.c144
-rw-r--r--tools/perf/tests/hists_link.c1
-rw-r--r--tools/perf/tests/make25
-rw-r--r--tools/perf/tests/parse-events.c2
-rw-r--r--tools/perf/tests/sample-parsing.c17
-rw-r--r--tools/perf/tests/tests.h9
-rw-r--r--tools/perf/ui/browsers/hists.c122
-rw-r--r--tools/perf/ui/gtk/hists.c78
-rw-r--r--tools/perf/ui/hist.c138
-rw-r--r--tools/perf/ui/stdio/hist.c11
-rw-r--r--tools/perf/util/annotate.c14
-rw-r--r--tools/perf/util/cpumap.c2
-rw-r--r--tools/perf/util/dso.c4
-rw-r--r--tools/perf/util/dso.h10
-rw-r--r--tools/perf/util/event.c188
-rw-r--r--tools/perf/util/event.h5
-rw-r--r--tools/perf/util/evsel.c60
-rw-r--r--tools/perf/util/evsel.h18
-rw-r--r--tools/perf/util/fs.h7
-rw-r--r--tools/perf/util/hist.c13
-rw-r--r--tools/perf/util/hist.h29
-rw-r--r--tools/perf/util/include/linux/hash.h5
-rw-r--r--tools/perf/util/include/linux/kernel.h6
-rw-r--r--tools/perf/util/include/linux/list.h1
-rw-r--r--tools/perf/util/include/linux/prefetch.h6
-rw-r--r--tools/perf/util/machine.c79
-rw-r--r--tools/perf/util/machine.h13
-rw-r--r--tools/perf/util/map.h10
-rw-r--r--tools/perf/util/parse-options.c37
-rw-r--r--tools/perf/util/parse-options.h8
-rw-r--r--tools/perf/util/perf_regs.c19
-rw-r--r--tools/perf/util/perf_regs.h13
-rw-r--r--tools/perf/util/pmu.c2
-rw-r--r--tools/perf/util/probe-event.c863
-rw-r--r--tools/perf/util/probe-event.h12
-rw-r--r--tools/perf/util/probe-finder.c198
-rw-r--r--tools/perf/util/probe-finder.h5
-rw-r--r--tools/perf/util/python-ext-sources2
-rw-r--r--tools/perf/util/record.c2
-rw-r--r--tools/perf/util/session.c7
-rw-r--r--tools/perf/util/symbol-elf.c2
-rw-r--r--tools/perf/util/symbol.c63
-rw-r--r--tools/perf/util/symbol.h14
-rw-r--r--tools/perf/util/thread.c21
-rw-r--r--tools/perf/util/thread.h11
-rw-r--r--tools/perf/util/trace-event-parse.c1
-rw-r--r--tools/perf/util/unwind-libdw.c210
-rw-r--r--tools/perf/util/unwind-libdw.h21
-rw-r--r--tools/perf/util/unwind-libunwind.c (renamed from tools/perf/util/unwind.c)50
-rw-r--r--tools/perf/util/unwind.h11
-rw-r--r--tools/perf/util/util.c2
102 files changed, 3472 insertions, 1289 deletions
diff --git a/arch/x86/include/asm/nmi.h b/arch/x86/include/asm/nmi.h
index 86f9301903c8..5f2fc4441b11 100644
--- a/arch/x86/include/asm/nmi.h
+++ b/arch/x86/include/asm/nmi.h
@@ -1,6 +1,7 @@
1#ifndef _ASM_X86_NMI_H 1#ifndef _ASM_X86_NMI_H
2#define _ASM_X86_NMI_H 2#define _ASM_X86_NMI_H
3 3
4#include <linux/irq_work.h>
4#include <linux/pm.h> 5#include <linux/pm.h>
5#include <asm/irq.h> 6#include <asm/irq.h>
6#include <asm/io.h> 7#include <asm/io.h>
@@ -38,6 +39,8 @@ typedef int (*nmi_handler_t)(unsigned int, struct pt_regs *);
38struct nmiaction { 39struct nmiaction {
39 struct list_head list; 40 struct list_head list;
40 nmi_handler_t handler; 41 nmi_handler_t handler;
42 u64 max_duration;
43 struct irq_work irq_work;
41 unsigned long flags; 44 unsigned long flags;
42 const char *name; 45 const char *name;
43}; 46};
diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c
index 79f9f848bee4..ae407f7226c8 100644
--- a/arch/x86/kernel/cpu/perf_event.c
+++ b/arch/x86/kernel/cpu/perf_event.c
@@ -892,7 +892,6 @@ static void x86_pmu_enable(struct pmu *pmu)
892 * hw_perf_group_sched_in() or x86_pmu_enable() 892 * hw_perf_group_sched_in() or x86_pmu_enable()
893 * 893 *
894 * step1: save events moving to new counters 894 * step1: save events moving to new counters
895 * step2: reprogram moved events into new counters
896 */ 895 */
897 for (i = 0; i < n_running; i++) { 896 for (i = 0; i < n_running; i++) {
898 event = cpuc->event_list[i]; 897 event = cpuc->event_list[i];
@@ -918,6 +917,9 @@ static void x86_pmu_enable(struct pmu *pmu)
918 x86_pmu_stop(event, PERF_EF_UPDATE); 917 x86_pmu_stop(event, PERF_EF_UPDATE);
919 } 918 }
920 919
920 /*
921 * step2: reprogram moved events into new counters
922 */
921 for (i = 0; i < cpuc->n_events; i++) { 923 for (i = 0; i < cpuc->n_events; i++) {
922 event = cpuc->event_list[i]; 924 event = cpuc->event_list[i];
923 hwc = &event->hw; 925 hwc = &event->hw;
@@ -1043,7 +1045,7 @@ static int x86_pmu_add(struct perf_event *event, int flags)
1043 /* 1045 /*
1044 * If group events scheduling transaction was started, 1046 * If group events scheduling transaction was started,
1045 * skip the schedulability test here, it will be performed 1047 * skip the schedulability test here, it will be performed
1046 * at commit time (->commit_txn) as a whole 1048 * at commit time (->commit_txn) as a whole.
1047 */ 1049 */
1048 if (cpuc->group_flag & PERF_EVENT_TXN) 1050 if (cpuc->group_flag & PERF_EVENT_TXN)
1049 goto done_collect; 1051 goto done_collect;
@@ -1058,6 +1060,10 @@ static int x86_pmu_add(struct perf_event *event, int flags)
1058 memcpy(cpuc->assign, assign, n*sizeof(int)); 1060 memcpy(cpuc->assign, assign, n*sizeof(int));
1059 1061
1060done_collect: 1062done_collect:
1063 /*
1064 * Commit the collect_events() state. See x86_pmu_del() and
1065 * x86_pmu_*_txn().
1066 */
1061 cpuc->n_events = n; 1067 cpuc->n_events = n;
1062 cpuc->n_added += n - n0; 1068 cpuc->n_added += n - n0;
1063 cpuc->n_txn += n - n0; 1069 cpuc->n_txn += n - n0;
@@ -1183,28 +1189,38 @@ static void x86_pmu_del(struct perf_event *event, int flags)
1183 * If we're called during a txn, we don't need to do anything. 1189 * If we're called during a txn, we don't need to do anything.
1184 * The events never got scheduled and ->cancel_txn will truncate 1190 * The events never got scheduled and ->cancel_txn will truncate
1185 * the event_list. 1191 * the event_list.
1192 *
1193 * XXX assumes any ->del() called during a TXN will only be on
1194 * an event added during that same TXN.
1186 */ 1195 */
1187 if (cpuc->group_flag & PERF_EVENT_TXN) 1196 if (cpuc->group_flag & PERF_EVENT_TXN)
1188 return; 1197 return;
1189 1198
1199 /*
1200 * Not a TXN, therefore cleanup properly.
1201 */
1190 x86_pmu_stop(event, PERF_EF_UPDATE); 1202 x86_pmu_stop(event, PERF_EF_UPDATE);
1191 1203
1192 for (i = 0; i < cpuc->n_events; i++) { 1204 for (i = 0; i < cpuc->n_events; i++) {
1193 if (event == cpuc->event_list[i]) { 1205 if (event == cpuc->event_list[i])
1206 break;
1207 }
1194 1208
1195 if (i >= cpuc->n_events - cpuc->n_added) 1209 if (WARN_ON_ONCE(i == cpuc->n_events)) /* called ->del() without ->add() ? */
1196 --cpuc->n_added; 1210 return;
1197 1211
1198 if (x86_pmu.put_event_constraints) 1212 /* If we have a newly added event; make sure to decrease n_added. */
1199 x86_pmu.put_event_constraints(cpuc, event); 1213 if (i >= cpuc->n_events - cpuc->n_added)
1214 --cpuc->n_added;
1200 1215
1201 while (++i < cpuc->n_events) 1216 if (x86_pmu.put_event_constraints)
1202 cpuc->event_list[i-1] = cpuc->event_list[i]; 1217 x86_pmu.put_event_constraints(cpuc, event);
1218
1219 /* Delete the array entry. */
1220 while (++i < cpuc->n_events)
1221 cpuc->event_list[i-1] = cpuc->event_list[i];
1222 --cpuc->n_events;
1203 1223
1204 --cpuc->n_events;
1205 break;
1206 }
1207 }
1208 perf_event_update_userpage(event); 1224 perf_event_update_userpage(event);
1209} 1225}
1210 1226
@@ -1598,7 +1614,8 @@ static void x86_pmu_cancel_txn(struct pmu *pmu)
1598{ 1614{
1599 __this_cpu_and(cpu_hw_events.group_flag, ~PERF_EVENT_TXN); 1615 __this_cpu_and(cpu_hw_events.group_flag, ~PERF_EVENT_TXN);
1600 /* 1616 /*
1601 * Truncate the collected events. 1617 * Truncate collected array by the number of events added in this
1618 * transaction. See x86_pmu_add() and x86_pmu_*_txn().
1602 */ 1619 */
1603 __this_cpu_sub(cpu_hw_events.n_added, __this_cpu_read(cpu_hw_events.n_txn)); 1620 __this_cpu_sub(cpu_hw_events.n_added, __this_cpu_read(cpu_hw_events.n_txn));
1604 __this_cpu_sub(cpu_hw_events.n_events, __this_cpu_read(cpu_hw_events.n_txn)); 1621 __this_cpu_sub(cpu_hw_events.n_events, __this_cpu_read(cpu_hw_events.n_txn));
@@ -1609,6 +1626,8 @@ static void x86_pmu_cancel_txn(struct pmu *pmu)
1609 * Commit group events scheduling transaction 1626 * Commit group events scheduling transaction
1610 * Perform the group schedulability test as a whole 1627 * Perform the group schedulability test as a whole
1611 * Return 0 if success 1628 * Return 0 if success
1629 *
1630 * Does not cancel the transaction on failure; expects the caller to do this.
1612 */ 1631 */
1613static int x86_pmu_commit_txn(struct pmu *pmu) 1632static int x86_pmu_commit_txn(struct pmu *pmu)
1614{ 1633{
diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h
index 4972c244d0bc..3b2f9bdd974b 100644
--- a/arch/x86/kernel/cpu/perf_event.h
+++ b/arch/x86/kernel/cpu/perf_event.h
@@ -130,9 +130,11 @@ struct cpu_hw_events {
130 unsigned long running[BITS_TO_LONGS(X86_PMC_IDX_MAX)]; 130 unsigned long running[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
131 int enabled; 131 int enabled;
132 132
133 int n_events; 133 int n_events; /* the # of events in the below arrays */
134 int n_added; 134 int n_added; /* the # last events in the below arrays;
135 int n_txn; 135 they've never been enabled yet */
136 int n_txn; /* the # last events in the below arrays;
137 added in the current transaction */
136 int assign[X86_PMC_IDX_MAX]; /* event to counter assignment */ 138 int assign[X86_PMC_IDX_MAX]; /* event to counter assignment */
137 u64 tags[X86_PMC_IDX_MAX]; 139 u64 tags[X86_PMC_IDX_MAX];
138 struct perf_event *event_list[X86_PMC_IDX_MAX]; /* in enabled order */ 140 struct perf_event *event_list[X86_PMC_IDX_MAX]; /* in enabled order */
diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.c b/arch/x86/kernel/cpu/perf_event_intel_uncore.c
index 047f540cf3f7..bd2253d40cff 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c
@@ -66,6 +66,47 @@ DEFINE_UNCORE_FORMAT_ATTR(mask_vnw, mask_vnw, "config2:3-4");
66DEFINE_UNCORE_FORMAT_ATTR(mask0, mask0, "config2:0-31"); 66DEFINE_UNCORE_FORMAT_ATTR(mask0, mask0, "config2:0-31");
67DEFINE_UNCORE_FORMAT_ATTR(mask1, mask1, "config2:32-63"); 67DEFINE_UNCORE_FORMAT_ATTR(mask1, mask1, "config2:32-63");
68 68
69static void uncore_pmu_start_hrtimer(struct intel_uncore_box *box);
70static void uncore_pmu_cancel_hrtimer(struct intel_uncore_box *box);
71static void uncore_perf_event_update(struct intel_uncore_box *box, struct perf_event *event);
72static void uncore_pmu_event_read(struct perf_event *event);
73
74static struct intel_uncore_pmu *uncore_event_to_pmu(struct perf_event *event)
75{
76 return container_of(event->pmu, struct intel_uncore_pmu, pmu);
77}
78
79static struct intel_uncore_box *
80uncore_pmu_to_box(struct intel_uncore_pmu *pmu, int cpu)
81{
82 struct intel_uncore_box *box;
83
84 box = *per_cpu_ptr(pmu->box, cpu);
85 if (box)
86 return box;
87
88 raw_spin_lock(&uncore_box_lock);
89 list_for_each_entry(box, &pmu->box_list, list) {
90 if (box->phys_id == topology_physical_package_id(cpu)) {
91 atomic_inc(&box->refcnt);
92 *per_cpu_ptr(pmu->box, cpu) = box;
93 break;
94 }
95 }
96 raw_spin_unlock(&uncore_box_lock);
97
98 return *per_cpu_ptr(pmu->box, cpu);
99}
100
101static struct intel_uncore_box *uncore_event_to_box(struct perf_event *event)
102{
103 /*
104 * perf core schedules event on the basis of cpu, uncore events are
105 * collected by one of the cpus inside a physical package.
106 */
107 return uncore_pmu_to_box(uncore_event_to_pmu(event), smp_processor_id());
108}
109
69static u64 uncore_msr_read_counter(struct intel_uncore_box *box, struct perf_event *event) 110static u64 uncore_msr_read_counter(struct intel_uncore_box *box, struct perf_event *event)
70{ 111{
71 u64 count; 112 u64 count;
@@ -1639,6 +1680,349 @@ static struct intel_uncore_type *snb_msr_uncores[] = {
1639 &snb_uncore_cbox, 1680 &snb_uncore_cbox,
1640 NULL, 1681 NULL,
1641}; 1682};
1683
1684enum {
1685 SNB_PCI_UNCORE_IMC,
1686};
1687
1688static struct uncore_event_desc snb_uncore_imc_events[] = {
1689 INTEL_UNCORE_EVENT_DESC(data_reads, "event=0x01"),
1690 INTEL_UNCORE_EVENT_DESC(data_reads.scale, "6.103515625e-5"),
1691 INTEL_UNCORE_EVENT_DESC(data_reads.unit, "MiB"),
1692
1693 INTEL_UNCORE_EVENT_DESC(data_writes, "event=0x02"),
1694 INTEL_UNCORE_EVENT_DESC(data_writes.scale, "6.103515625e-5"),
1695 INTEL_UNCORE_EVENT_DESC(data_writes.unit, "MiB"),
1696
1697 { /* end: all zeroes */ },
1698};
1699
1700#define SNB_UNCORE_PCI_IMC_EVENT_MASK 0xff
1701#define SNB_UNCORE_PCI_IMC_BAR_OFFSET 0x48
1702
1703/* page size multiple covering all config regs */
1704#define SNB_UNCORE_PCI_IMC_MAP_SIZE 0x6000
1705
1706#define SNB_UNCORE_PCI_IMC_DATA_READS 0x1
1707#define SNB_UNCORE_PCI_IMC_DATA_READS_BASE 0x5050
1708#define SNB_UNCORE_PCI_IMC_DATA_WRITES 0x2
1709#define SNB_UNCORE_PCI_IMC_DATA_WRITES_BASE 0x5054
1710#define SNB_UNCORE_PCI_IMC_CTR_BASE SNB_UNCORE_PCI_IMC_DATA_READS_BASE
1711
1712static struct attribute *snb_uncore_imc_formats_attr[] = {
1713 &format_attr_event.attr,
1714 NULL,
1715};
1716
1717static struct attribute_group snb_uncore_imc_format_group = {
1718 .name = "format",
1719 .attrs = snb_uncore_imc_formats_attr,
1720};
1721
1722static void snb_uncore_imc_init_box(struct intel_uncore_box *box)
1723{
1724 struct pci_dev *pdev = box->pci_dev;
1725 int where = SNB_UNCORE_PCI_IMC_BAR_OFFSET;
1726 resource_size_t addr;
1727 u32 pci_dword;
1728
1729 pci_read_config_dword(pdev, where, &pci_dword);
1730 addr = pci_dword;
1731
1732#ifdef CONFIG_PHYS_ADDR_T_64BIT
1733 pci_read_config_dword(pdev, where + 4, &pci_dword);
1734 addr |= ((resource_size_t)pci_dword << 32);
1735#endif
1736
1737 addr &= ~(PAGE_SIZE - 1);
1738
1739 box->io_addr = ioremap(addr, SNB_UNCORE_PCI_IMC_MAP_SIZE);
1740 box->hrtimer_duration = UNCORE_SNB_IMC_HRTIMER_INTERVAL;
1741}
1742
1743static void snb_uncore_imc_enable_box(struct intel_uncore_box *box)
1744{}
1745
1746static void snb_uncore_imc_disable_box(struct intel_uncore_box *box)
1747{}
1748
1749static void snb_uncore_imc_enable_event(struct intel_uncore_box *box, struct perf_event *event)
1750{}
1751
1752static void snb_uncore_imc_disable_event(struct intel_uncore_box *box, struct perf_event *event)
1753{}
1754
1755static u64 snb_uncore_imc_read_counter(struct intel_uncore_box *box, struct perf_event *event)
1756{
1757 struct hw_perf_event *hwc = &event->hw;
1758
1759 return (u64)*(unsigned int *)(box->io_addr + hwc->event_base);
1760}
1761
1762/*
1763 * custom event_init() function because we define our own fixed, free
1764 * running counters, so we do not want to conflict with generic uncore
1765 * logic. Also simplifies processing
1766 */
1767static int snb_uncore_imc_event_init(struct perf_event *event)
1768{
1769 struct intel_uncore_pmu *pmu;
1770 struct intel_uncore_box *box;
1771 struct hw_perf_event *hwc = &event->hw;
1772 u64 cfg = event->attr.config & SNB_UNCORE_PCI_IMC_EVENT_MASK;
1773 int idx, base;
1774
1775 if (event->attr.type != event->pmu->type)
1776 return -ENOENT;
1777
1778 pmu = uncore_event_to_pmu(event);
1779 /* no device found for this pmu */
1780 if (pmu->func_id < 0)
1781 return -ENOENT;
1782
1783 /* Sampling not supported yet */
1784 if (hwc->sample_period)
1785 return -EINVAL;
1786
1787 /* unsupported modes and filters */
1788 if (event->attr.exclude_user ||
1789 event->attr.exclude_kernel ||
1790 event->attr.exclude_hv ||
1791 event->attr.exclude_idle ||
1792 event->attr.exclude_host ||
1793 event->attr.exclude_guest ||
1794 event->attr.sample_period) /* no sampling */
1795 return -EINVAL;
1796
1797 /*
1798 * Place all uncore events for a particular physical package
1799 * onto a single cpu
1800 */
1801 if (event->cpu < 0)
1802 return -EINVAL;
1803
1804 /* check only supported bits are set */
1805 if (event->attr.config & ~SNB_UNCORE_PCI_IMC_EVENT_MASK)
1806 return -EINVAL;
1807
1808 box = uncore_pmu_to_box(pmu, event->cpu);
1809 if (!box || box->cpu < 0)
1810 return -EINVAL;
1811
1812 event->cpu = box->cpu;
1813
1814 event->hw.idx = -1;
1815 event->hw.last_tag = ~0ULL;
1816 event->hw.extra_reg.idx = EXTRA_REG_NONE;
1817 event->hw.branch_reg.idx = EXTRA_REG_NONE;
1818 /*
1819 * check event is known (whitelist, determines counter)
1820 */
1821 switch (cfg) {
1822 case SNB_UNCORE_PCI_IMC_DATA_READS:
1823 base = SNB_UNCORE_PCI_IMC_DATA_READS_BASE;
1824 idx = UNCORE_PMC_IDX_FIXED;
1825 break;
1826 case SNB_UNCORE_PCI_IMC_DATA_WRITES:
1827 base = SNB_UNCORE_PCI_IMC_DATA_WRITES_BASE;
1828 idx = UNCORE_PMC_IDX_FIXED + 1;
1829 break;
1830 default:
1831 return -EINVAL;
1832 }
1833
1834 /* must be done before validate_group */
1835 event->hw.event_base = base;
1836 event->hw.config = cfg;
1837 event->hw.idx = idx;
1838
1839 /* no group validation needed, we have free running counters */
1840
1841 return 0;
1842}
1843
1844static int snb_uncore_imc_hw_config(struct intel_uncore_box *box, struct perf_event *event)
1845{
1846 return 0;
1847}
1848
1849static void snb_uncore_imc_event_start(struct perf_event *event, int flags)
1850{
1851 struct intel_uncore_box *box = uncore_event_to_box(event);
1852 u64 count;
1853
1854 if (WARN_ON_ONCE(!(event->hw.state & PERF_HES_STOPPED)))
1855 return;
1856
1857 event->hw.state = 0;
1858 box->n_active++;
1859
1860 list_add_tail(&event->active_entry, &box->active_list);
1861
1862 count = snb_uncore_imc_read_counter(box, event);
1863 local64_set(&event->hw.prev_count, count);
1864
1865 if (box->n_active == 1)
1866 uncore_pmu_start_hrtimer(box);
1867}
1868
1869static void snb_uncore_imc_event_stop(struct perf_event *event, int flags)
1870{
1871 struct intel_uncore_box *box = uncore_event_to_box(event);
1872 struct hw_perf_event *hwc = &event->hw;
1873
1874 if (!(hwc->state & PERF_HES_STOPPED)) {
1875 box->n_active--;
1876
1877 WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
1878 hwc->state |= PERF_HES_STOPPED;
1879
1880 list_del(&event->active_entry);
1881
1882 if (box->n_active == 0)
1883 uncore_pmu_cancel_hrtimer(box);
1884 }
1885
1886 if ((flags & PERF_EF_UPDATE) && !(hwc->state & PERF_HES_UPTODATE)) {
1887 /*
1888 * Drain the remaining delta count out of a event
1889 * that we are disabling:
1890 */
1891 uncore_perf_event_update(box, event);
1892 hwc->state |= PERF_HES_UPTODATE;
1893 }
1894}
1895
1896static int snb_uncore_imc_event_add(struct perf_event *event, int flags)
1897{
1898 struct intel_uncore_box *box = uncore_event_to_box(event);
1899 struct hw_perf_event *hwc = &event->hw;
1900
1901 if (!box)
1902 return -ENODEV;
1903
1904 hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
1905 if (!(flags & PERF_EF_START))
1906 hwc->state |= PERF_HES_ARCH;
1907
1908 snb_uncore_imc_event_start(event, 0);
1909
1910 box->n_events++;
1911
1912 return 0;
1913}
1914
1915static void snb_uncore_imc_event_del(struct perf_event *event, int flags)
1916{
1917 struct intel_uncore_box *box = uncore_event_to_box(event);
1918 int i;
1919
1920 snb_uncore_imc_event_stop(event, PERF_EF_UPDATE);
1921
1922 for (i = 0; i < box->n_events; i++) {
1923 if (event == box->event_list[i]) {
1924 --box->n_events;
1925 break;
1926 }
1927 }
1928}
1929
1930static int snb_pci2phy_map_init(int devid)
1931{
1932 struct pci_dev *dev = NULL;
1933 int bus;
1934
1935 dev = pci_get_device(PCI_VENDOR_ID_INTEL, devid, dev);
1936 if (!dev)
1937 return -ENOTTY;
1938
1939 bus = dev->bus->number;
1940
1941 pcibus_to_physid[bus] = 0;
1942
1943 pci_dev_put(dev);
1944
1945 return 0;
1946}
1947
1948static struct pmu snb_uncore_imc_pmu = {
1949 .task_ctx_nr = perf_invalid_context,
1950 .event_init = snb_uncore_imc_event_init,
1951 .add = snb_uncore_imc_event_add,
1952 .del = snb_uncore_imc_event_del,
1953 .start = snb_uncore_imc_event_start,
1954 .stop = snb_uncore_imc_event_stop,
1955 .read = uncore_pmu_event_read,
1956};
1957
1958static struct intel_uncore_ops snb_uncore_imc_ops = {
1959 .init_box = snb_uncore_imc_init_box,
1960 .enable_box = snb_uncore_imc_enable_box,
1961 .disable_box = snb_uncore_imc_disable_box,
1962 .disable_event = snb_uncore_imc_disable_event,
1963 .enable_event = snb_uncore_imc_enable_event,
1964 .hw_config = snb_uncore_imc_hw_config,
1965 .read_counter = snb_uncore_imc_read_counter,
1966};
1967
1968static struct intel_uncore_type snb_uncore_imc = {
1969 .name = "imc",
1970 .num_counters = 2,
1971 .num_boxes = 1,
1972 .fixed_ctr_bits = 32,
1973 .fixed_ctr = SNB_UNCORE_PCI_IMC_CTR_BASE,
1974 .event_descs = snb_uncore_imc_events,
1975 .format_group = &snb_uncore_imc_format_group,
1976 .perf_ctr = SNB_UNCORE_PCI_IMC_DATA_READS_BASE,
1977 .event_mask = SNB_UNCORE_PCI_IMC_EVENT_MASK,
1978 .ops = &snb_uncore_imc_ops,
1979 .pmu = &snb_uncore_imc_pmu,
1980};
1981
1982static struct intel_uncore_type *snb_pci_uncores[] = {
1983 [SNB_PCI_UNCORE_IMC] = &snb_uncore_imc,
1984 NULL,
1985};
1986
1987static DEFINE_PCI_DEVICE_TABLE(snb_uncore_pci_ids) = {
1988 { /* IMC */
1989 PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_SNB_IMC),
1990 .driver_data = UNCORE_PCI_DEV_DATA(SNB_PCI_UNCORE_IMC, 0),
1991 },
1992 { /* end: all zeroes */ },
1993};
1994
1995static DEFINE_PCI_DEVICE_TABLE(ivb_uncore_pci_ids) = {
1996 { /* IMC */
1997 PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IVB_IMC),
1998 .driver_data = UNCORE_PCI_DEV_DATA(SNB_PCI_UNCORE_IMC, 0),
1999 },
2000 { /* end: all zeroes */ },
2001};
2002
2003static DEFINE_PCI_DEVICE_TABLE(hsw_uncore_pci_ids) = {
2004 { /* IMC */
2005 PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_HSW_IMC),
2006 .driver_data = UNCORE_PCI_DEV_DATA(SNB_PCI_UNCORE_IMC, 0),
2007 },
2008 { /* end: all zeroes */ },
2009};
2010
2011static struct pci_driver snb_uncore_pci_driver = {
2012 .name = "snb_uncore",
2013 .id_table = snb_uncore_pci_ids,
2014};
2015
2016static struct pci_driver ivb_uncore_pci_driver = {
2017 .name = "ivb_uncore",
2018 .id_table = ivb_uncore_pci_ids,
2019};
2020
2021static struct pci_driver hsw_uncore_pci_driver = {
2022 .name = "hsw_uncore",
2023 .id_table = hsw_uncore_pci_ids,
2024};
2025
1642/* end of Sandy Bridge uncore support */ 2026/* end of Sandy Bridge uncore support */
1643 2027
1644/* Nehalem uncore support */ 2028/* Nehalem uncore support */
@@ -2789,6 +3173,7 @@ again:
2789static enum hrtimer_restart uncore_pmu_hrtimer(struct hrtimer *hrtimer) 3173static enum hrtimer_restart uncore_pmu_hrtimer(struct hrtimer *hrtimer)
2790{ 3174{
2791 struct intel_uncore_box *box; 3175 struct intel_uncore_box *box;
3176 struct perf_event *event;
2792 unsigned long flags; 3177 unsigned long flags;
2793 int bit; 3178 int bit;
2794 3179
@@ -2801,19 +3186,27 @@ static enum hrtimer_restart uncore_pmu_hrtimer(struct hrtimer *hrtimer)
2801 */ 3186 */
2802 local_irq_save(flags); 3187 local_irq_save(flags);
2803 3188
3189 /*
3190 * handle boxes with an active event list as opposed to active
3191 * counters
3192 */
3193 list_for_each_entry(event, &box->active_list, active_entry) {
3194 uncore_perf_event_update(box, event);
3195 }
3196
2804 for_each_set_bit(bit, box->active_mask, UNCORE_PMC_IDX_MAX) 3197 for_each_set_bit(bit, box->active_mask, UNCORE_PMC_IDX_MAX)
2805 uncore_perf_event_update(box, box->events[bit]); 3198 uncore_perf_event_update(box, box->events[bit]);
2806 3199
2807 local_irq_restore(flags); 3200 local_irq_restore(flags);
2808 3201
2809 hrtimer_forward_now(hrtimer, ns_to_ktime(UNCORE_PMU_HRTIMER_INTERVAL)); 3202 hrtimer_forward_now(hrtimer, ns_to_ktime(box->hrtimer_duration));
2810 return HRTIMER_RESTART; 3203 return HRTIMER_RESTART;
2811} 3204}
2812 3205
2813static void uncore_pmu_start_hrtimer(struct intel_uncore_box *box) 3206static void uncore_pmu_start_hrtimer(struct intel_uncore_box *box)
2814{ 3207{
2815 __hrtimer_start_range_ns(&box->hrtimer, 3208 __hrtimer_start_range_ns(&box->hrtimer,
2816 ns_to_ktime(UNCORE_PMU_HRTIMER_INTERVAL), 0, 3209 ns_to_ktime(box->hrtimer_duration), 0,
2817 HRTIMER_MODE_REL_PINNED, 0); 3210 HRTIMER_MODE_REL_PINNED, 0);
2818} 3211}
2819 3212
@@ -2847,43 +3240,12 @@ static struct intel_uncore_box *uncore_alloc_box(struct intel_uncore_type *type,
2847 box->cpu = -1; 3240 box->cpu = -1;
2848 box->phys_id = -1; 3241 box->phys_id = -1;
2849 3242
2850 return box; 3243 /* set default hrtimer timeout */
2851} 3244 box->hrtimer_duration = UNCORE_PMU_HRTIMER_INTERVAL;
2852 3245
2853static struct intel_uncore_box * 3246 INIT_LIST_HEAD(&box->active_list);
2854uncore_pmu_to_box(struct intel_uncore_pmu *pmu, int cpu)
2855{
2856 struct intel_uncore_box *box;
2857 3247
2858 box = *per_cpu_ptr(pmu->box, cpu); 3248 return box;
2859 if (box)
2860 return box;
2861
2862 raw_spin_lock(&uncore_box_lock);
2863 list_for_each_entry(box, &pmu->box_list, list) {
2864 if (box->phys_id == topology_physical_package_id(cpu)) {
2865 atomic_inc(&box->refcnt);
2866 *per_cpu_ptr(pmu->box, cpu) = box;
2867 break;
2868 }
2869 }
2870 raw_spin_unlock(&uncore_box_lock);
2871
2872 return *per_cpu_ptr(pmu->box, cpu);
2873}
2874
2875static struct intel_uncore_pmu *uncore_event_to_pmu(struct perf_event *event)
2876{
2877 return container_of(event->pmu, struct intel_uncore_pmu, pmu);
2878}
2879
2880static struct intel_uncore_box *uncore_event_to_box(struct perf_event *event)
2881{
2882 /*
2883 * perf core schedules event on the basis of cpu, uncore events are
2884 * collected by one of the cpus inside a physical package.
2885 */
2886 return uncore_pmu_to_box(uncore_event_to_pmu(event), smp_processor_id());
2887} 3249}
2888 3250
2889static int 3251static int
@@ -3279,16 +3641,21 @@ static int __init uncore_pmu_register(struct intel_uncore_pmu *pmu)
3279{ 3641{
3280 int ret; 3642 int ret;
3281 3643
3282 pmu->pmu = (struct pmu) { 3644 if (!pmu->type->pmu) {
3283 .attr_groups = pmu->type->attr_groups, 3645 pmu->pmu = (struct pmu) {
3284 .task_ctx_nr = perf_invalid_context, 3646 .attr_groups = pmu->type->attr_groups,
3285 .event_init = uncore_pmu_event_init, 3647 .task_ctx_nr = perf_invalid_context,
3286 .add = uncore_pmu_event_add, 3648 .event_init = uncore_pmu_event_init,
3287 .del = uncore_pmu_event_del, 3649 .add = uncore_pmu_event_add,
3288 .start = uncore_pmu_event_start, 3650 .del = uncore_pmu_event_del,
3289 .stop = uncore_pmu_event_stop, 3651 .start = uncore_pmu_event_start,
3290 .read = uncore_pmu_event_read, 3652 .stop = uncore_pmu_event_stop,
3291 }; 3653 .read = uncore_pmu_event_read,
3654 };
3655 } else {
3656 pmu->pmu = *pmu->type->pmu;
3657 pmu->pmu.attr_groups = pmu->type->attr_groups;
3658 }
3292 3659
3293 if (pmu->type->num_boxes == 1) { 3660 if (pmu->type->num_boxes == 1) {
3294 if (strlen(pmu->type->name) > 0) 3661 if (strlen(pmu->type->name) > 0)
@@ -3502,6 +3869,28 @@ static int __init uncore_pci_init(void)
3502 pci_uncores = ivt_pci_uncores; 3869 pci_uncores = ivt_pci_uncores;
3503 uncore_pci_driver = &ivt_uncore_pci_driver; 3870 uncore_pci_driver = &ivt_uncore_pci_driver;
3504 break; 3871 break;
3872 case 42: /* Sandy Bridge */
3873 ret = snb_pci2phy_map_init(PCI_DEVICE_ID_INTEL_SNB_IMC);
3874 if (ret)
3875 return ret;
3876 pci_uncores = snb_pci_uncores;
3877 uncore_pci_driver = &snb_uncore_pci_driver;
3878 break;
3879 case 58: /* Ivy Bridge */
3880 ret = snb_pci2phy_map_init(PCI_DEVICE_ID_INTEL_IVB_IMC);
3881 if (ret)
3882 return ret;
3883 pci_uncores = snb_pci_uncores;
3884 uncore_pci_driver = &ivb_uncore_pci_driver;
3885 break;
3886 case 60: /* Haswell */
3887 case 69: /* Haswell Celeron */
3888 ret = snb_pci2phy_map_init(PCI_DEVICE_ID_INTEL_HSW_IMC);
3889 if (ret)
3890 return ret;
3891 pci_uncores = snb_pci_uncores;
3892 uncore_pci_driver = &hsw_uncore_pci_driver;
3893 break;
3505 default: 3894 default:
3506 return 0; 3895 return 0;
3507 } 3896 }
@@ -3773,7 +4162,7 @@ static void __init uncore_cpu_setup(void *dummy)
3773 4162
3774static int __init uncore_cpu_init(void) 4163static int __init uncore_cpu_init(void)
3775{ 4164{
3776 int ret, cpu, max_cores; 4165 int ret, max_cores;
3777 4166
3778 max_cores = boot_cpu_data.x86_max_cores; 4167 max_cores = boot_cpu_data.x86_max_cores;
3779 switch (boot_cpu_data.x86_model) { 4168 switch (boot_cpu_data.x86_model) {
@@ -3817,29 +4206,6 @@ static int __init uncore_cpu_init(void)
3817 if (ret) 4206 if (ret)
3818 return ret; 4207 return ret;
3819 4208
3820 get_online_cpus();
3821
3822 for_each_online_cpu(cpu) {
3823 int i, phys_id = topology_physical_package_id(cpu);
3824
3825 for_each_cpu(i, &uncore_cpu_mask) {
3826 if (phys_id == topology_physical_package_id(i)) {
3827 phys_id = -1;
3828 break;
3829 }
3830 }
3831 if (phys_id < 0)
3832 continue;
3833
3834 uncore_cpu_prepare(cpu, phys_id);
3835 uncore_event_init_cpu(cpu);
3836 }
3837 on_each_cpu(uncore_cpu_setup, NULL, 1);
3838
3839 register_cpu_notifier(&uncore_cpu_nb);
3840
3841 put_online_cpus();
3842
3843 return 0; 4209 return 0;
3844} 4210}
3845 4211
@@ -3868,6 +4234,41 @@ static int __init uncore_pmus_register(void)
3868 return 0; 4234 return 0;
3869} 4235}
3870 4236
4237static void __init uncore_cpumask_init(void)
4238{
4239 int cpu;
4240
4241 /*
4242 * ony invoke once from msr or pci init code
4243 */
4244 if (!cpumask_empty(&uncore_cpu_mask))
4245 return;
4246
4247 get_online_cpus();
4248
4249 for_each_online_cpu(cpu) {
4250 int i, phys_id = topology_physical_package_id(cpu);
4251
4252 for_each_cpu(i, &uncore_cpu_mask) {
4253 if (phys_id == topology_physical_package_id(i)) {
4254 phys_id = -1;
4255 break;
4256 }
4257 }
4258 if (phys_id < 0)
4259 continue;
4260
4261 uncore_cpu_prepare(cpu, phys_id);
4262 uncore_event_init_cpu(cpu);
4263 }
4264 on_each_cpu(uncore_cpu_setup, NULL, 1);
4265
4266 register_cpu_notifier(&uncore_cpu_nb);
4267
4268 put_online_cpus();
4269}
4270
4271
3871static int __init intel_uncore_init(void) 4272static int __init intel_uncore_init(void)
3872{ 4273{
3873 int ret; 4274 int ret;
@@ -3886,6 +4287,7 @@ static int __init intel_uncore_init(void)
3886 uncore_pci_exit(); 4287 uncore_pci_exit();
3887 goto fail; 4288 goto fail;
3888 } 4289 }
4290 uncore_cpumask_init();
3889 4291
3890 uncore_pmus_register(); 4292 uncore_pmus_register();
3891 return 0; 4293 return 0;
diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.h b/arch/x86/kernel/cpu/perf_event_intel_uncore.h
index a80ab71a883d..90236f0c94a9 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_uncore.h
+++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.h
@@ -6,6 +6,7 @@
6 6
7#define UNCORE_PMU_NAME_LEN 32 7#define UNCORE_PMU_NAME_LEN 32
8#define UNCORE_PMU_HRTIMER_INTERVAL (60LL * NSEC_PER_SEC) 8#define UNCORE_PMU_HRTIMER_INTERVAL (60LL * NSEC_PER_SEC)
9#define UNCORE_SNB_IMC_HRTIMER_INTERVAL (5ULL * NSEC_PER_SEC)
9 10
10#define UNCORE_FIXED_EVENT 0xff 11#define UNCORE_FIXED_EVENT 0xff
11#define UNCORE_PMC_IDX_MAX_GENERIC 8 12#define UNCORE_PMC_IDX_MAX_GENERIC 8
@@ -440,6 +441,7 @@ struct intel_uncore_type {
440 struct intel_uncore_ops *ops; 441 struct intel_uncore_ops *ops;
441 struct uncore_event_desc *event_descs; 442 struct uncore_event_desc *event_descs;
442 const struct attribute_group *attr_groups[4]; 443 const struct attribute_group *attr_groups[4];
444 struct pmu *pmu; /* for custom pmu ops */
443}; 445};
444 446
445#define pmu_group attr_groups[0] 447#define pmu_group attr_groups[0]
@@ -488,8 +490,11 @@ struct intel_uncore_box {
488 u64 tags[UNCORE_PMC_IDX_MAX]; 490 u64 tags[UNCORE_PMC_IDX_MAX];
489 struct pci_dev *pci_dev; 491 struct pci_dev *pci_dev;
490 struct intel_uncore_pmu *pmu; 492 struct intel_uncore_pmu *pmu;
493 u64 hrtimer_duration; /* hrtimer timeout for this box */
491 struct hrtimer hrtimer; 494 struct hrtimer hrtimer;
492 struct list_head list; 495 struct list_head list;
496 struct list_head active_list;
497 void *io_addr;
493 struct intel_uncore_extra_reg shared_regs[0]; 498 struct intel_uncore_extra_reg shared_regs[0];
494}; 499};
495 500
diff --git a/arch/x86/kernel/cpu/perf_event_p4.c b/arch/x86/kernel/cpu/perf_event_p4.c
index 3486e6660357..5d466b7d8609 100644
--- a/arch/x86/kernel/cpu/perf_event_p4.c
+++ b/arch/x86/kernel/cpu/perf_event_p4.c
@@ -1257,7 +1257,24 @@ again:
1257 pass++; 1257 pass++;
1258 goto again; 1258 goto again;
1259 } 1259 }
1260 1260 /*
1261 * Perf does test runs to see if a whole group can be assigned
1262 * together succesfully. There can be multiple rounds of this.
1263 * Unfortunately, p4_pmu_swap_config_ts touches the hwc->config
1264 * bits, such that the next round of group assignments will
1265 * cause the above p4_should_swap_ts to pass instead of fail.
1266 * This leads to counters exclusive to thread0 being used by
1267 * thread1.
1268 *
1269 * Solve this with a cheap hack, reset the idx back to -1 to
1270 * force a new lookup (p4_next_cntr) to get the right counter
1271 * for the right thread.
1272 *
1273 * This probably doesn't comply with the general spirit of how
1274 * perf wants to work, but P4 is special. :-(
1275 */
1276 if (p4_should_swap_ts(hwc->config, cpu))
1277 hwc->idx = -1;
1261 p4_pmu_swap_config_ts(hwc, cpu); 1278 p4_pmu_swap_config_ts(hwc, cpu);
1262 if (assign) 1279 if (assign)
1263 assign[i] = cntr_idx; 1280 assign[i] = cntr_idx;
@@ -1322,6 +1339,7 @@ static __initconst const struct x86_pmu p4_pmu = {
1322__init int p4_pmu_init(void) 1339__init int p4_pmu_init(void)
1323{ 1340{
1324 unsigned int low, high; 1341 unsigned int low, high;
1342 int i, reg;
1325 1343
1326 /* If we get stripped -- indexing fails */ 1344 /* If we get stripped -- indexing fails */
1327 BUILD_BUG_ON(ARCH_P4_MAX_CCCR > INTEL_PMC_MAX_GENERIC); 1345 BUILD_BUG_ON(ARCH_P4_MAX_CCCR > INTEL_PMC_MAX_GENERIC);
@@ -1340,5 +1358,19 @@ __init int p4_pmu_init(void)
1340 1358
1341 x86_pmu = p4_pmu; 1359 x86_pmu = p4_pmu;
1342 1360
1361 /*
1362 * Even though the counters are configured to interrupt a particular
1363 * logical processor when an overflow happens, testing has shown that
1364 * on kdump kernels (which uses a single cpu), thread1's counter
1365 * continues to run and will report an NMI on thread0. Due to the
1366 * overflow bug, this leads to a stream of unknown NMIs.
1367 *
1368 * Solve this by zero'ing out the registers to mimic a reset.
1369 */
1370 for (i = 0; i < x86_pmu.num_counters; i++) {
1371 reg = x86_pmu_config_addr(i);
1372 wrmsrl_safe(reg, 0ULL);
1373 }
1374
1343 return 0; 1375 return 0;
1344} 1376}
diff --git a/arch/x86/kernel/nmi.c b/arch/x86/kernel/nmi.c
index 6fcb49ce50a1..b4872b999a71 100644
--- a/arch/x86/kernel/nmi.c
+++ b/arch/x86/kernel/nmi.c
@@ -87,6 +87,7 @@ __setup("unknown_nmi_panic", setup_unknown_nmi_panic);
87#define nmi_to_desc(type) (&nmi_desc[type]) 87#define nmi_to_desc(type) (&nmi_desc[type])
88 88
89static u64 nmi_longest_ns = 1 * NSEC_PER_MSEC; 89static u64 nmi_longest_ns = 1 * NSEC_PER_MSEC;
90
90static int __init nmi_warning_debugfs(void) 91static int __init nmi_warning_debugfs(void)
91{ 92{
92 debugfs_create_u64("nmi_longest_ns", 0644, 93 debugfs_create_u64("nmi_longest_ns", 0644,
@@ -95,6 +96,20 @@ static int __init nmi_warning_debugfs(void)
95} 96}
96fs_initcall(nmi_warning_debugfs); 97fs_initcall(nmi_warning_debugfs);
97 98
99static void nmi_max_handler(struct irq_work *w)
100{
101 struct nmiaction *a = container_of(w, struct nmiaction, irq_work);
102 int remainder_ns, decimal_msecs;
103 u64 whole_msecs = ACCESS_ONCE(a->max_duration);
104
105 remainder_ns = do_div(whole_msecs, (1000 * 1000));
106 decimal_msecs = remainder_ns / 1000;
107
108 printk_ratelimited(KERN_INFO
109 "INFO: NMI handler (%ps) took too long to run: %lld.%03d msecs\n",
110 a->handler, whole_msecs, decimal_msecs);
111}
112
98static int __kprobes nmi_handle(unsigned int type, struct pt_regs *regs, bool b2b) 113static int __kprobes nmi_handle(unsigned int type, struct pt_regs *regs, bool b2b)
99{ 114{
100 struct nmi_desc *desc = nmi_to_desc(type); 115 struct nmi_desc *desc = nmi_to_desc(type);
@@ -110,26 +125,20 @@ static int __kprobes nmi_handle(unsigned int type, struct pt_regs *regs, bool b2
110 * to handle those situations. 125 * to handle those situations.
111 */ 126 */
112 list_for_each_entry_rcu(a, &desc->head, list) { 127 list_for_each_entry_rcu(a, &desc->head, list) {
113 u64 before, delta, whole_msecs; 128 int thishandled;
114 int remainder_ns, decimal_msecs, thishandled; 129 u64 delta;
115 130
116 before = sched_clock(); 131 delta = sched_clock();
117 thishandled = a->handler(type, regs); 132 thishandled = a->handler(type, regs);
118 handled += thishandled; 133 handled += thishandled;
119 delta = sched_clock() - before; 134 delta = sched_clock() - delta;
120 trace_nmi_handler(a->handler, (int)delta, thishandled); 135 trace_nmi_handler(a->handler, (int)delta, thishandled);
121 136
122 if (delta < nmi_longest_ns) 137 if (delta < nmi_longest_ns || delta < a->max_duration)
123 continue; 138 continue;
124 139
125 nmi_longest_ns = delta; 140 a->max_duration = delta;
126 whole_msecs = delta; 141 irq_work_queue(&a->irq_work);
127 remainder_ns = do_div(whole_msecs, (1000 * 1000));
128 decimal_msecs = remainder_ns / 1000;
129 printk_ratelimited(KERN_INFO
130 "INFO: NMI handler (%ps) took too long to run: "
131 "%lld.%03d msecs\n", a->handler, whole_msecs,
132 decimal_msecs);
133 } 142 }
134 143
135 rcu_read_unlock(); 144 rcu_read_unlock();
@@ -146,6 +155,8 @@ int __register_nmi_handler(unsigned int type, struct nmiaction *action)
146 if (!action->handler) 155 if (!action->handler)
147 return -EINVAL; 156 return -EINVAL;
148 157
158 init_irq_work(&action->irq_work, nmi_max_handler);
159
149 spin_lock_irqsave(&desc->lock, flags); 160 spin_lock_irqsave(&desc->lock, flags);
150 161
151 /* 162 /*
diff --git a/include/linux/irq_work.h b/include/linux/irq_work.h
index 66017028dcb3..19ae05d4b8ec 100644
--- a/include/linux/irq_work.h
+++ b/include/linux/irq_work.h
@@ -30,7 +30,9 @@ void init_irq_work(struct irq_work *work, void (*func)(struct irq_work *))
30 work->func = func; 30 work->func = func;
31} 31}
32 32
33void irq_work_queue(struct irq_work *work); 33#define DEFINE_IRQ_WORK(name, _f) struct irq_work name = { .func = (_f), }
34
35bool irq_work_queue(struct irq_work *work);
34void irq_work_run(void); 36void irq_work_run(void);
35void irq_work_sync(struct irq_work *work); 37void irq_work_sync(struct irq_work *work);
36 38
diff --git a/include/linux/pci_ids.h b/include/linux/pci_ids.h
index 97fbecdd7a40..7399e6a3e9a0 100644
--- a/include/linux/pci_ids.h
+++ b/include/linux/pci_ids.h
@@ -2531,6 +2531,9 @@
2531 2531
2532#define PCI_VENDOR_ID_INTEL 0x8086 2532#define PCI_VENDOR_ID_INTEL 0x8086
2533#define PCI_DEVICE_ID_INTEL_EESSC 0x0008 2533#define PCI_DEVICE_ID_INTEL_EESSC 0x0008
2534#define PCI_DEVICE_ID_INTEL_SNB_IMC 0x0100
2535#define PCI_DEVICE_ID_INTEL_IVB_IMC 0x0154
2536#define PCI_DEVICE_ID_INTEL_HSW_IMC 0x0c00
2534#define PCI_DEVICE_ID_INTEL_PXHD_0 0x0320 2537#define PCI_DEVICE_ID_INTEL_PXHD_0 0x0320
2535#define PCI_DEVICE_ID_INTEL_PXHD_1 0x0321 2538#define PCI_DEVICE_ID_INTEL_PXHD_1 0x0321
2536#define PCI_DEVICE_ID_INTEL_PXH_0 0x0329 2539#define PCI_DEVICE_ID_INTEL_PXH_0 0x0329
diff --git a/kernel/events/core.c b/kernel/events/core.c
index fa0b2d4ad83c..661951ab8ae7 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -231,11 +231,29 @@ int perf_cpu_time_max_percent_handler(struct ctl_table *table, int write,
231#define NR_ACCUMULATED_SAMPLES 128 231#define NR_ACCUMULATED_SAMPLES 128
232static DEFINE_PER_CPU(u64, running_sample_length); 232static DEFINE_PER_CPU(u64, running_sample_length);
233 233
234void perf_sample_event_took(u64 sample_len_ns) 234static void perf_duration_warn(struct irq_work *w)
235{ 235{
236 u64 allowed_ns = ACCESS_ONCE(perf_sample_allowed_ns);
236 u64 avg_local_sample_len; 237 u64 avg_local_sample_len;
237 u64 local_samples_len; 238 u64 local_samples_len;
239
240 local_samples_len = __get_cpu_var(running_sample_length);
241 avg_local_sample_len = local_samples_len/NR_ACCUMULATED_SAMPLES;
242
243 printk_ratelimited(KERN_WARNING
244 "perf interrupt took too long (%lld > %lld), lowering "
245 "kernel.perf_event_max_sample_rate to %d\n",
246 avg_local_sample_len, allowed_ns >> 1,
247 sysctl_perf_event_sample_rate);
248}
249
250static DEFINE_IRQ_WORK(perf_duration_work, perf_duration_warn);
251
252void perf_sample_event_took(u64 sample_len_ns)
253{
238 u64 allowed_ns = ACCESS_ONCE(perf_sample_allowed_ns); 254 u64 allowed_ns = ACCESS_ONCE(perf_sample_allowed_ns);
255 u64 avg_local_sample_len;
256 u64 local_samples_len;
239 257
240 if (allowed_ns == 0) 258 if (allowed_ns == 0)
241 return; 259 return;
@@ -263,13 +281,14 @@ void perf_sample_event_took(u64 sample_len_ns)
263 sysctl_perf_event_sample_rate = max_samples_per_tick * HZ; 281 sysctl_perf_event_sample_rate = max_samples_per_tick * HZ;
264 perf_sample_period_ns = NSEC_PER_SEC / sysctl_perf_event_sample_rate; 282 perf_sample_period_ns = NSEC_PER_SEC / sysctl_perf_event_sample_rate;
265 283
266 printk_ratelimited(KERN_WARNING
267 "perf samples too long (%lld > %lld), lowering "
268 "kernel.perf_event_max_sample_rate to %d\n",
269 avg_local_sample_len, allowed_ns,
270 sysctl_perf_event_sample_rate);
271
272 update_perf_cpu_limits(); 284 update_perf_cpu_limits();
285
286 if (!irq_work_queue(&perf_duration_work)) {
287 early_printk("perf interrupt took too long (%lld > %lld), lowering "
288 "kernel.perf_event_max_sample_rate to %d\n",
289 avg_local_sample_len, allowed_ns >> 1,
290 sysctl_perf_event_sample_rate);
291 }
273} 292}
274 293
275static atomic64_t perf_event_id; 294static atomic64_t perf_event_id;
@@ -1714,7 +1733,7 @@ group_sched_in(struct perf_event *group_event,
1714 struct perf_event_context *ctx) 1733 struct perf_event_context *ctx)
1715{ 1734{
1716 struct perf_event *event, *partial_group = NULL; 1735 struct perf_event *event, *partial_group = NULL;
1717 struct pmu *pmu = group_event->pmu; 1736 struct pmu *pmu = ctx->pmu;
1718 u64 now = ctx->time; 1737 u64 now = ctx->time;
1719 bool simulate = false; 1738 bool simulate = false;
1720 1739
@@ -2563,8 +2582,6 @@ static void perf_branch_stack_sched_in(struct task_struct *prev,
2563 if (cpuctx->ctx.nr_branch_stack > 0 2582 if (cpuctx->ctx.nr_branch_stack > 0
2564 && pmu->flush_branch_stack) { 2583 && pmu->flush_branch_stack) {
2565 2584
2566 pmu = cpuctx->ctx.pmu;
2567
2568 perf_ctx_lock(cpuctx, cpuctx->task_ctx); 2585 perf_ctx_lock(cpuctx, cpuctx->task_ctx);
2569 2586
2570 perf_pmu_disable(pmu); 2587 perf_pmu_disable(pmu);
@@ -6294,7 +6311,7 @@ static int perf_event_idx_default(struct perf_event *event)
6294 * Ensures all contexts with the same task_ctx_nr have the same 6311 * Ensures all contexts with the same task_ctx_nr have the same
6295 * pmu_cpu_context too. 6312 * pmu_cpu_context too.
6296 */ 6313 */
6297static void *find_pmu_context(int ctxn) 6314static struct perf_cpu_context __percpu *find_pmu_context(int ctxn)
6298{ 6315{
6299 struct pmu *pmu; 6316 struct pmu *pmu;
6300 6317
diff --git a/kernel/irq_work.c b/kernel/irq_work.c
index 55fcce6065cf..a82170e2fa78 100644
--- a/kernel/irq_work.c
+++ b/kernel/irq_work.c
@@ -61,11 +61,11 @@ void __weak arch_irq_work_raise(void)
61 * 61 *
62 * Can be re-enqueued while the callback is still in progress. 62 * Can be re-enqueued while the callback is still in progress.
63 */ 63 */
64void irq_work_queue(struct irq_work *work) 64bool irq_work_queue(struct irq_work *work)
65{ 65{
66 /* Only queue if not already pending */ 66 /* Only queue if not already pending */
67 if (!irq_work_claim(work)) 67 if (!irq_work_claim(work))
68 return; 68 return false;
69 69
70 /* Queue the entry and raise the IPI if needed. */ 70 /* Queue the entry and raise the IPI if needed. */
71 preempt_disable(); 71 preempt_disable();
@@ -83,6 +83,8 @@ void irq_work_queue(struct irq_work *work)
83 } 83 }
84 84
85 preempt_enable(); 85 preempt_enable();
86
87 return true;
86} 88}
87EXPORT_SYMBOL_GPL(irq_work_queue); 89EXPORT_SYMBOL_GPL(irq_work_queue);
88 90
diff --git a/kernel/trace/trace_event_perf.c b/kernel/trace/trace_event_perf.c
index e854f420e033..c894614de14d 100644
--- a/kernel/trace/trace_event_perf.c
+++ b/kernel/trace/trace_event_perf.c
@@ -31,9 +31,25 @@ static int perf_trace_event_perm(struct ftrace_event_call *tp_event,
31 } 31 }
32 32
33 /* The ftrace function trace is allowed only for root. */ 33 /* The ftrace function trace is allowed only for root. */
34 if (ftrace_event_is_function(tp_event) && 34 if (ftrace_event_is_function(tp_event)) {
35 perf_paranoid_tracepoint_raw() && !capable(CAP_SYS_ADMIN)) 35 if (perf_paranoid_tracepoint_raw() && !capable(CAP_SYS_ADMIN))
36 return -EPERM; 36 return -EPERM;
37
38 /*
39 * We don't allow user space callchains for function trace
40 * event, due to issues with page faults while tracing page
41 * fault handler and its overall trickiness nature.
42 */
43 if (!p_event->attr.exclude_callchain_user)
44 return -EINVAL;
45
46 /*
47 * Same reason to disable user stack dump as for user space
48 * callchains above.
49 */
50 if (p_event->attr.sample_type & PERF_SAMPLE_STACK_USER)
51 return -EINVAL;
52 }
37 53
38 /* No tracing, just counting, so no obvious leak */ 54 /* No tracing, just counting, so no obvious leak */
39 if (!(p_event->attr.sample_type & PERF_SAMPLE_RAW)) 55 if (!(p_event->attr.sample_type & PERF_SAMPLE_RAW))
diff --git a/tools/include/linux/hash.h b/tools/include/linux/hash.h
new file mode 100644
index 000000000000..d026c6573018
--- /dev/null
+++ b/tools/include/linux/hash.h
@@ -0,0 +1,5 @@
1#include "../../../include/linux/hash.h"
2
3#ifndef _TOOLS_LINUX_HASH_H
4#define _TOOLS_LINUX_HASH_H
5#endif
diff --git a/tools/lib/api/Makefile b/tools/lib/api/Makefile
index ed2f51e11b80..ce00f7ee6455 100644
--- a/tools/lib/api/Makefile
+++ b/tools/lib/api/Makefile
@@ -9,8 +9,10 @@ LIB_H=
9LIB_OBJS= 9LIB_OBJS=
10 10
11LIB_H += fs/debugfs.h 11LIB_H += fs/debugfs.h
12LIB_H += fs/fs.h
12 13
13LIB_OBJS += $(OUTPUT)fs/debugfs.o 14LIB_OBJS += $(OUTPUT)fs/debugfs.o
15LIB_OBJS += $(OUTPUT)fs/fs.o
14 16
15LIBFILE = libapikfs.a 17LIBFILE = libapikfs.a
16 18
diff --git a/tools/perf/util/fs.c b/tools/lib/api/fs/fs.c
index f5be1f26e724..5b5eb788996e 100644
--- a/tools/perf/util/fs.c
+++ b/tools/lib/api/fs/fs.c
@@ -1,8 +1,13 @@
1/* TODO merge/factor in debugfs.c here */
1 2
2/* TODO merge/factor into tools/lib/lk/debugfs.c */ 3#include <errno.h>
4#include <stdbool.h>
5#include <stdio.h>
6#include <string.h>
7#include <sys/vfs.h>
3 8
4#include "util.h" 9#include "debugfs.h"
5#include "util/fs.h" 10#include "fs.h"
6 11
7static const char * const sysfs__fs_known_mountpoints[] = { 12static const char * const sysfs__fs_known_mountpoints[] = {
8 "/sys", 13 "/sys",
diff --git a/tools/perf/util/include/linux/magic.h b/tools/lib/api/fs/fs.h
index 07d63cf3e0f6..cb7049551f33 100644
--- a/tools/perf/util/include/linux/magic.h
+++ b/tools/lib/api/fs/fs.h
@@ -1,9 +1,5 @@
1#ifndef _PERF_LINUX_MAGIC_H_ 1#ifndef __API_FS__
2#define _PERF_LINUX_MAGIC_H_ 2#define __API_FS__
3
4#ifndef DEBUGFS_MAGIC
5#define DEBUGFS_MAGIC 0x64626720
6#endif
7 3
8#ifndef SYSFS_MAGIC 4#ifndef SYSFS_MAGIC
9#define SYSFS_MAGIC 0x62656572 5#define SYSFS_MAGIC 0x62656572
@@ -13,4 +9,6 @@
13#define PROC_SUPER_MAGIC 0x9fa0 9#define PROC_SUPER_MAGIC 0x9fa0
14#endif 10#endif
15 11
16#endif 12const char *sysfs__mountpoint(void);
13const char *procfs__mountpoint(void);
14#endif /* __API_FS__ */
diff --git a/tools/perf/Documentation/perf-mem.txt b/tools/perf/Documentation/perf-mem.txt
index 888d51137fbe..1d78a4064da4 100644
--- a/tools/perf/Documentation/perf-mem.txt
+++ b/tools/perf/Documentation/perf-mem.txt
@@ -18,6 +18,10 @@ from it, into perf.data. Perf record options are accepted and are passed through
18"perf mem -t <TYPE> report" displays the result. It invokes perf report with the 18"perf mem -t <TYPE> report" displays the result. It invokes perf report with the
19right set of options to display a memory access profile. 19right set of options to display a memory access profile.
20 20
21Note that on Intel systems the memory latency reported is the use-latency,
22not the pure load (or store latency). Use latency includes any pipeline
23queueing delays in addition to the memory subsystem latency.
24
21OPTIONS 25OPTIONS
22------- 26-------
23<command>...:: 27<command>...::
diff --git a/tools/perf/Documentation/perf-probe.txt b/tools/perf/Documentation/perf-probe.txt
index b715cb71592b..1513935c399b 100644
--- a/tools/perf/Documentation/perf-probe.txt
+++ b/tools/perf/Documentation/perf-probe.txt
@@ -136,6 +136,8 @@ Each probe argument follows below syntax.
136'NAME' specifies the name of this argument (optional). You can use the name of local variable, local data structure member (e.g. var->field, var.field2), local array with fixed index (e.g. array[1], var->array[0], var->pointer[2]), or kprobe-tracer argument format (e.g. $retval, %ax, etc). Note that the name of this argument will be set as the last member name if you specify a local data structure member (e.g. field2 for 'var->field1.field2'.) 136'NAME' specifies the name of this argument (optional). You can use the name of local variable, local data structure member (e.g. var->field, var.field2), local array with fixed index (e.g. array[1], var->array[0], var->pointer[2]), or kprobe-tracer argument format (e.g. $retval, %ax, etc). Note that the name of this argument will be set as the last member name if you specify a local data structure member (e.g. field2 for 'var->field1.field2'.)
137'TYPE' casts the type of this argument (optional). If omitted, perf probe automatically set the type based on debuginfo. You can specify 'string' type only for the local variable or structure member which is an array of or a pointer to 'char' or 'unsigned char' type. 137'TYPE' casts the type of this argument (optional). If omitted, perf probe automatically set the type based on debuginfo. You can specify 'string' type only for the local variable or structure member which is an array of or a pointer to 'char' or 'unsigned char' type.
138 138
139On x86 systems %REG is always the short form of the register: for example %AX. %RAX or %EAX is not valid.
140
139LINE SYNTAX 141LINE SYNTAX
140----------- 142-----------
141Line range is described by following syntax. 143Line range is described by following syntax.
diff --git a/tools/perf/MANIFEST b/tools/perf/MANIFEST
index f41572d0dd76..c0c87c87b60f 100644
--- a/tools/perf/MANIFEST
+++ b/tools/perf/MANIFEST
@@ -6,6 +6,7 @@ tools/lib/symbol/kallsyms.c
6tools/lib/symbol/kallsyms.h 6tools/lib/symbol/kallsyms.h
7tools/include/asm/bug.h 7tools/include/asm/bug.h
8tools/include/linux/compiler.h 8tools/include/linux/compiler.h
9tools/include/linux/hash.h
9include/linux/const.h 10include/linux/const.h
10include/linux/perf_event.h 11include/linux/perf_event.h
11include/linux/rbtree.h 12include/linux/rbtree.h
diff --git a/tools/perf/Makefile.perf b/tools/perf/Makefile.perf
index 7257e7e9e38a..50d875d970c4 100644
--- a/tools/perf/Makefile.perf
+++ b/tools/perf/Makefile.perf
@@ -7,6 +7,8 @@ include config/utilities.mak
7 7
8# Define V to have a more verbose compile. 8# Define V to have a more verbose compile.
9# 9#
10# Define VF to have a more verbose feature check output.
11#
10# Define O to save output files in a separate directory. 12# Define O to save output files in a separate directory.
11# 13#
12# Define ARCH as name of target architecture if you want cross-builds. 14# Define ARCH as name of target architecture if you want cross-builds.
@@ -55,6 +57,9 @@ include config/utilities.mak
55# Define NO_LIBAUDIT if you do not want libaudit support 57# Define NO_LIBAUDIT if you do not want libaudit support
56# 58#
57# Define NO_LIBBIONIC if you do not want bionic support 59# Define NO_LIBBIONIC if you do not want bionic support
60#
61# Define NO_LIBDW_DWARF_UNWIND if you do not want libdw support
62# for dwarf backtrace post unwind.
58 63
59ifeq ($(srctree),) 64ifeq ($(srctree),)
60srctree := $(patsubst %/,%,$(dir $(shell pwd))) 65srctree := $(patsubst %/,%,$(dir $(shell pwd)))
@@ -208,7 +213,7 @@ LIB_H += ../../include/uapi/linux/perf_event.h
208LIB_H += ../../include/linux/rbtree.h 213LIB_H += ../../include/linux/rbtree.h
209LIB_H += ../../include/linux/list.h 214LIB_H += ../../include/linux/list.h
210LIB_H += ../../include/uapi/linux/const.h 215LIB_H += ../../include/uapi/linux/const.h
211LIB_H += ../../include/linux/hash.h 216LIB_H += ../include/linux/hash.h
212LIB_H += ../../include/linux/stringify.h 217LIB_H += ../../include/linux/stringify.h
213LIB_H += util/include/linux/bitmap.h 218LIB_H += util/include/linux/bitmap.h
214LIB_H += util/include/linux/bitops.h 219LIB_H += util/include/linux/bitops.h
@@ -218,9 +223,7 @@ LIB_H += util/include/linux/ctype.h
218LIB_H += util/include/linux/kernel.h 223LIB_H += util/include/linux/kernel.h
219LIB_H += util/include/linux/list.h 224LIB_H += util/include/linux/list.h
220LIB_H += util/include/linux/export.h 225LIB_H += util/include/linux/export.h
221LIB_H += util/include/linux/magic.h
222LIB_H += util/include/linux/poison.h 226LIB_H += util/include/linux/poison.h
223LIB_H += util/include/linux/prefetch.h
224LIB_H += util/include/linux/rbtree.h 227LIB_H += util/include/linux/rbtree.h
225LIB_H += util/include/linux/rbtree_augmented.h 228LIB_H += util/include/linux/rbtree_augmented.h
226LIB_H += util/include/linux/string.h 229LIB_H += util/include/linux/string.h
@@ -244,7 +247,6 @@ LIB_H += util/cache.h
244LIB_H += util/callchain.h 247LIB_H += util/callchain.h
245LIB_H += util/build-id.h 248LIB_H += util/build-id.h
246LIB_H += util/debug.h 249LIB_H += util/debug.h
247LIB_H += util/fs.h
248LIB_H += util/pmu.h 250LIB_H += util/pmu.h
249LIB_H += util/event.h 251LIB_H += util/event.h
250LIB_H += util/evsel.h 252LIB_H += util/evsel.h
@@ -306,7 +308,6 @@ LIB_OBJS += $(OUTPUT)util/annotate.o
306LIB_OBJS += $(OUTPUT)util/build-id.o 308LIB_OBJS += $(OUTPUT)util/build-id.o
307LIB_OBJS += $(OUTPUT)util/config.o 309LIB_OBJS += $(OUTPUT)util/config.o
308LIB_OBJS += $(OUTPUT)util/ctype.o 310LIB_OBJS += $(OUTPUT)util/ctype.o
309LIB_OBJS += $(OUTPUT)util/fs.o
310LIB_OBJS += $(OUTPUT)util/pmu.o 311LIB_OBJS += $(OUTPUT)util/pmu.o
311LIB_OBJS += $(OUTPUT)util/environment.o 312LIB_OBJS += $(OUTPUT)util/environment.o
312LIB_OBJS += $(OUTPUT)util/event.o 313LIB_OBJS += $(OUTPUT)util/event.o
@@ -408,6 +409,11 @@ endif
408LIB_OBJS += $(OUTPUT)tests/code-reading.o 409LIB_OBJS += $(OUTPUT)tests/code-reading.o
409LIB_OBJS += $(OUTPUT)tests/sample-parsing.o 410LIB_OBJS += $(OUTPUT)tests/sample-parsing.o
410LIB_OBJS += $(OUTPUT)tests/parse-no-sample-id-all.o 411LIB_OBJS += $(OUTPUT)tests/parse-no-sample-id-all.o
412ifndef NO_DWARF_UNWIND
413ifeq ($(ARCH),x86)
414LIB_OBJS += $(OUTPUT)tests/dwarf-unwind.o
415endif
416endif
411 417
412BUILTIN_OBJS += $(OUTPUT)builtin-annotate.o 418BUILTIN_OBJS += $(OUTPUT)builtin-annotate.o
413BUILTIN_OBJS += $(OUTPUT)builtin-bench.o 419BUILTIN_OBJS += $(OUTPUT)builtin-bench.o
@@ -420,6 +426,9 @@ BUILTIN_OBJS += $(OUTPUT)bench/mem-memset-x86-64-asm.o
420endif 426endif
421BUILTIN_OBJS += $(OUTPUT)bench/mem-memcpy.o 427BUILTIN_OBJS += $(OUTPUT)bench/mem-memcpy.o
422BUILTIN_OBJS += $(OUTPUT)bench/mem-memset.o 428BUILTIN_OBJS += $(OUTPUT)bench/mem-memset.o
429BUILTIN_OBJS += $(OUTPUT)bench/futex-hash.o
430BUILTIN_OBJS += $(OUTPUT)bench/futex-wake.o
431BUILTIN_OBJS += $(OUTPUT)bench/futex-requeue.o
423 432
424BUILTIN_OBJS += $(OUTPUT)builtin-diff.o 433BUILTIN_OBJS += $(OUTPUT)builtin-diff.o
425BUILTIN_OBJS += $(OUTPUT)builtin-evlist.o 434BUILTIN_OBJS += $(OUTPUT)builtin-evlist.o
@@ -475,8 +484,13 @@ ifndef NO_DWARF
475endif # NO_DWARF 484endif # NO_DWARF
476endif # NO_LIBELF 485endif # NO_LIBELF
477 486
487ifndef NO_LIBDW_DWARF_UNWIND
488 LIB_OBJS += $(OUTPUT)util/unwind-libdw.o
489 LIB_H += util/unwind-libdw.h
490endif
491
478ifndef NO_LIBUNWIND 492ifndef NO_LIBUNWIND
479 LIB_OBJS += $(OUTPUT)util/unwind.o 493 LIB_OBJS += $(OUTPUT)util/unwind-libunwind.o
480endif 494endif
481LIB_OBJS += $(OUTPUT)tests/keep-tracking.o 495LIB_OBJS += $(OUTPUT)tests/keep-tracking.o
482 496
@@ -533,6 +547,7 @@ ifeq ($(NO_PERF_REGS),0)
533 ifeq ($(ARCH),x86) 547 ifeq ($(ARCH),x86)
534 LIB_H += arch/x86/include/perf_regs.h 548 LIB_H += arch/x86/include/perf_regs.h
535 endif 549 endif
550 LIB_OBJS += $(OUTPUT)util/perf_regs.o
536endif 551endif
537 552
538ifndef NO_LIBNUMA 553ifndef NO_LIBNUMA
@@ -655,6 +670,9 @@ $(OUTPUT)tests/python-use.o: tests/python-use.c $(OUTPUT)PERF-CFLAGS
655 -DPYTHON='"$(PYTHON_WORD)"' \ 670 -DPYTHON='"$(PYTHON_WORD)"' \
656 $< 671 $<
657 672
673$(OUTPUT)tests/dwarf-unwind.o: tests/dwarf-unwind.c
674 $(QUIET_CC)$(CC) -o $@ -c $(CFLAGS) -fno-optimize-sibling-calls $<
675
658$(OUTPUT)util/config.o: util/config.c $(OUTPUT)PERF-CFLAGS 676$(OUTPUT)util/config.o: util/config.c $(OUTPUT)PERF-CFLAGS
659 $(QUIET_CC)$(CC) -o $@ -c $(CFLAGS) -DETC_PERFCONFIG='"$(ETC_PERFCONFIG_SQ)"' $< 677 $(QUIET_CC)$(CC) -o $@ -c $(CFLAGS) -DETC_PERFCONFIG='"$(ETC_PERFCONFIG_SQ)"' $<
660 678
@@ -707,9 +725,15 @@ $(patsubst perf-%,%.o,$(PROGRAMS)): $(LIB_H) $(wildcard */*.h)
707# we depend the various files onto their directories. 725# we depend the various files onto their directories.
708DIRECTORY_DEPS = $(LIB_OBJS) $(BUILTIN_OBJS) $(GTK_OBJS) 726DIRECTORY_DEPS = $(LIB_OBJS) $(BUILTIN_OBJS) $(GTK_OBJS)
709DIRECTORY_DEPS += $(OUTPUT)PERF-VERSION-FILE $(OUTPUT)common-cmds.h 727DIRECTORY_DEPS += $(OUTPUT)PERF-VERSION-FILE $(OUTPUT)common-cmds.h
710$(DIRECTORY_DEPS): | $(sort $(dir $(DIRECTORY_DEPS))) 728# no need to add flex objects, because they depend on bison ones
729DIRECTORY_DEPS += $(OUTPUT)util/parse-events-bison.c
730DIRECTORY_DEPS += $(OUTPUT)util/pmu-bison.c
731
732OUTPUT_DIRECTORIES := $(sort $(dir $(DIRECTORY_DEPS)))
733
734$(DIRECTORY_DEPS): | $(OUTPUT_DIRECTORIES)
711# In the second step, we make a rule to actually create these directories 735# In the second step, we make a rule to actually create these directories
712$(sort $(dir $(DIRECTORY_DEPS))): 736$(OUTPUT_DIRECTORIES):
713 $(QUIET_MKDIR)$(MKDIR) -p $@ 2>/dev/null 737 $(QUIET_MKDIR)$(MKDIR) -p $@ 2>/dev/null
714 738
715$(LIB_FILE): $(LIB_OBJS) 739$(LIB_FILE): $(LIB_OBJS)
@@ -886,7 +910,7 @@ config-clean:
886clean: $(LIBTRACEEVENT)-clean $(LIBAPIKFS)-clean config-clean 910clean: $(LIBTRACEEVENT)-clean $(LIBAPIKFS)-clean config-clean
887 $(call QUIET_CLEAN, core-objs) $(RM) $(LIB_OBJS) $(BUILTIN_OBJS) $(LIB_FILE) $(OUTPUT)perf-archive $(OUTPUT)perf.o $(LANG_BINDINGS) $(GTK_OBJS) 911 $(call QUIET_CLEAN, core-objs) $(RM) $(LIB_OBJS) $(BUILTIN_OBJS) $(LIB_FILE) $(OUTPUT)perf-archive $(OUTPUT)perf.o $(LANG_BINDINGS) $(GTK_OBJS)
888 $(call QUIET_CLEAN, core-progs) $(RM) $(ALL_PROGRAMS) perf 912 $(call QUIET_CLEAN, core-progs) $(RM) $(ALL_PROGRAMS) perf
889 $(call QUIET_CLEAN, core-gen) $(RM) *.spec *.pyc *.pyo */*.pyc */*.pyo $(OUTPUT)common-cmds.h TAGS tags cscope* $(OUTPUT)PERF-VERSION-FILE $(OUTPUT)PERF-CFLAGS $(OUTPUT)util/*-bison* $(OUTPUT)util/*-flex* 913 $(call QUIET_CLEAN, core-gen) $(RM) *.spec *.pyc *.pyo */*.pyc */*.pyo $(OUTPUT)common-cmds.h TAGS tags cscope* $(OUTPUT)PERF-VERSION-FILE $(OUTPUT)PERF-CFLAGS $(OUTPUT)PERF-FEATURES $(OUTPUT)util/*-bison* $(OUTPUT)util/*-flex*
890 $(QUIET_SUBDIR0)Documentation $(QUIET_SUBDIR1) clean 914 $(QUIET_SUBDIR0)Documentation $(QUIET_SUBDIR1) clean
891 $(python-clean) 915 $(python-clean)
892 916
diff --git a/tools/perf/arch/arm/Makefile b/tools/perf/arch/arm/Makefile
index fe9b61e322a5..67e9b3d38e89 100644
--- a/tools/perf/arch/arm/Makefile
+++ b/tools/perf/arch/arm/Makefile
@@ -3,5 +3,5 @@ PERF_HAVE_DWARF_REGS := 1
3LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/dwarf-regs.o 3LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/dwarf-regs.o
4endif 4endif
5ifndef NO_LIBUNWIND 5ifndef NO_LIBUNWIND
6LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/unwind.o 6LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/unwind-libunwind.o
7endif 7endif
diff --git a/tools/perf/arch/arm/util/unwind.c b/tools/perf/arch/arm/util/unwind-libunwind.c
index da3dc950550c..729ed69a6664 100644
--- a/tools/perf/arch/arm/util/unwind.c
+++ b/tools/perf/arch/arm/util/unwind-libunwind.c
@@ -4,7 +4,7 @@
4#include "perf_regs.h" 4#include "perf_regs.h"
5#include "../../util/unwind.h" 5#include "../../util/unwind.h"
6 6
7int unwind__arch_reg_id(int regnum) 7int libunwind__arch_reg_id(int regnum)
8{ 8{
9 switch (regnum) { 9 switch (regnum) {
10 case UNW_ARM_R0: 10 case UNW_ARM_R0:
diff --git a/tools/perf/arch/x86/Makefile b/tools/perf/arch/x86/Makefile
index 8801fe02f206..1641542e3636 100644
--- a/tools/perf/arch/x86/Makefile
+++ b/tools/perf/arch/x86/Makefile
@@ -3,7 +3,14 @@ PERF_HAVE_DWARF_REGS := 1
3LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/dwarf-regs.o 3LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/dwarf-regs.o
4endif 4endif
5ifndef NO_LIBUNWIND 5ifndef NO_LIBUNWIND
6LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/unwind.o 6LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/unwind-libunwind.o
7endif
8ifndef NO_LIBDW_DWARF_UNWIND
9LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/unwind-libdw.o
10endif
11ifndef NO_DWARF_UNWIND
12LIB_OBJS += $(OUTPUT)arch/$(ARCH)/tests/regs_load.o
13LIB_OBJS += $(OUTPUT)arch/$(ARCH)/tests/dwarf-unwind.o
7endif 14endif
8LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/header.o 15LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/header.o
9LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/tsc.o 16LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/tsc.o
diff --git a/tools/perf/arch/x86/include/perf_regs.h b/tools/perf/arch/x86/include/perf_regs.h
index e84ca76aae77..fc819ca34a7e 100644
--- a/tools/perf/arch/x86/include/perf_regs.h
+++ b/tools/perf/arch/x86/include/perf_regs.h
@@ -5,14 +5,20 @@
5#include "../../util/types.h" 5#include "../../util/types.h"
6#include <asm/perf_regs.h> 6#include <asm/perf_regs.h>
7 7
8void perf_regs_load(u64 *regs);
9
8#ifndef HAVE_ARCH_X86_64_SUPPORT 10#ifndef HAVE_ARCH_X86_64_SUPPORT
9#define PERF_REGS_MASK ((1ULL << PERF_REG_X86_32_MAX) - 1) 11#define PERF_REGS_MASK ((1ULL << PERF_REG_X86_32_MAX) - 1)
12#define PERF_REGS_MAX PERF_REG_X86_32_MAX
13#define PERF_SAMPLE_REGS_ABI PERF_SAMPLE_REGS_ABI_32
10#else 14#else
11#define REG_NOSUPPORT ((1ULL << PERF_REG_X86_DS) | \ 15#define REG_NOSUPPORT ((1ULL << PERF_REG_X86_DS) | \
12 (1ULL << PERF_REG_X86_ES) | \ 16 (1ULL << PERF_REG_X86_ES) | \
13 (1ULL << PERF_REG_X86_FS) | \ 17 (1ULL << PERF_REG_X86_FS) | \
14 (1ULL << PERF_REG_X86_GS)) 18 (1ULL << PERF_REG_X86_GS))
15#define PERF_REGS_MASK (((1ULL << PERF_REG_X86_64_MAX) - 1) & ~REG_NOSUPPORT) 19#define PERF_REGS_MASK (((1ULL << PERF_REG_X86_64_MAX) - 1) & ~REG_NOSUPPORT)
20#define PERF_REGS_MAX PERF_REG_X86_64_MAX
21#define PERF_SAMPLE_REGS_ABI PERF_SAMPLE_REGS_ABI_64
16#endif 22#endif
17#define PERF_REG_IP PERF_REG_X86_IP 23#define PERF_REG_IP PERF_REG_X86_IP
18#define PERF_REG_SP PERF_REG_X86_SP 24#define PERF_REG_SP PERF_REG_X86_SP
diff --git a/tools/perf/arch/x86/tests/dwarf-unwind.c b/tools/perf/arch/x86/tests/dwarf-unwind.c
new file mode 100644
index 000000000000..b602ad93ce63
--- /dev/null
+++ b/tools/perf/arch/x86/tests/dwarf-unwind.c
@@ -0,0 +1,59 @@
1#include <string.h>
2#include "perf_regs.h"
3#include "thread.h"
4#include "map.h"
5#include "event.h"
6#include "tests/tests.h"
7
8#define STACK_SIZE 8192
9
10static int sample_ustack(struct perf_sample *sample,
11 struct thread *thread, u64 *regs)
12{
13 struct stack_dump *stack = &sample->user_stack;
14 struct map *map;
15 unsigned long sp;
16 u64 stack_size, *buf;
17
18 buf = malloc(STACK_SIZE);
19 if (!buf) {
20 pr_debug("failed to allocate sample uregs data\n");
21 return -1;
22 }
23
24 sp = (unsigned long) regs[PERF_REG_X86_SP];
25
26 map = map_groups__find(&thread->mg, MAP__FUNCTION, (u64) sp);
27 if (!map) {
28 pr_debug("failed to get stack map\n");
29 return -1;
30 }
31
32 stack_size = map->end - sp;
33 stack_size = stack_size > STACK_SIZE ? STACK_SIZE : stack_size;
34
35 memcpy(buf, (void *) sp, stack_size);
36 stack->data = (char *) buf;
37 stack->size = stack_size;
38 return 0;
39}
40
41int test__arch_unwind_sample(struct perf_sample *sample,
42 struct thread *thread)
43{
44 struct regs_dump *regs = &sample->user_regs;
45 u64 *buf;
46
47 buf = malloc(sizeof(u64) * PERF_REGS_MAX);
48 if (!buf) {
49 pr_debug("failed to allocate sample uregs data\n");
50 return -1;
51 }
52
53 perf_regs_load(buf);
54 regs->abi = PERF_SAMPLE_REGS_ABI;
55 regs->regs = buf;
56 regs->mask = PERF_REGS_MASK;
57
58 return sample_ustack(sample, thread, buf);
59}
diff --git a/tools/perf/arch/x86/tests/regs_load.S b/tools/perf/arch/x86/tests/regs_load.S
new file mode 100644
index 000000000000..99167bf644ea
--- /dev/null
+++ b/tools/perf/arch/x86/tests/regs_load.S
@@ -0,0 +1,92 @@
1
2#include <linux/linkage.h>
3
4#define AX 0
5#define BX 1 * 8
6#define CX 2 * 8
7#define DX 3 * 8
8#define SI 4 * 8
9#define DI 5 * 8
10#define BP 6 * 8
11#define SP 7 * 8
12#define IP 8 * 8
13#define FLAGS 9 * 8
14#define CS 10 * 8
15#define SS 11 * 8
16#define DS 12 * 8
17#define ES 13 * 8
18#define FS 14 * 8
19#define GS 15 * 8
20#define R8 16 * 8
21#define R9 17 * 8
22#define R10 18 * 8
23#define R11 19 * 8
24#define R12 20 * 8
25#define R13 21 * 8
26#define R14 22 * 8
27#define R15 23 * 8
28
29.text
30#ifdef HAVE_ARCH_X86_64_SUPPORT
31ENTRY(perf_regs_load)
32 movq %rax, AX(%rdi)
33 movq %rbx, BX(%rdi)
34 movq %rcx, CX(%rdi)
35 movq %rdx, DX(%rdi)
36 movq %rsi, SI(%rdi)
37 movq %rdi, DI(%rdi)
38 movq %rbp, BP(%rdi)
39
40 leaq 8(%rsp), %rax /* exclude this call. */
41 movq %rax, SP(%rdi)
42
43 movq 0(%rsp), %rax
44 movq %rax, IP(%rdi)
45
46 movq $0, FLAGS(%rdi)
47 movq $0, CS(%rdi)
48 movq $0, SS(%rdi)
49 movq $0, DS(%rdi)
50 movq $0, ES(%rdi)
51 movq $0, FS(%rdi)
52 movq $0, GS(%rdi)
53
54 movq %r8, R8(%rdi)
55 movq %r9, R9(%rdi)
56 movq %r10, R10(%rdi)
57 movq %r11, R11(%rdi)
58 movq %r12, R12(%rdi)
59 movq %r13, R13(%rdi)
60 movq %r14, R14(%rdi)
61 movq %r15, R15(%rdi)
62 ret
63ENDPROC(perf_regs_load)
64#else
65ENTRY(perf_regs_load)
66 push %edi
67 movl 8(%esp), %edi
68 movl %eax, AX(%edi)
69 movl %ebx, BX(%edi)
70 movl %ecx, CX(%edi)
71 movl %edx, DX(%edi)
72 movl %esi, SI(%edi)
73 pop %eax
74 movl %eax, DI(%edi)
75 movl %ebp, BP(%edi)
76
77 leal 4(%esp), %eax /* exclude this call. */
78 movl %eax, SP(%edi)
79
80 movl 0(%esp), %eax
81 movl %eax, IP(%edi)
82
83 movl $0, FLAGS(%edi)
84 movl $0, CS(%edi)
85 movl $0, SS(%edi)
86 movl $0, DS(%edi)
87 movl $0, ES(%edi)
88 movl $0, FS(%edi)
89 movl $0, GS(%edi)
90 ret
91ENDPROC(perf_regs_load)
92#endif
diff --git a/tools/perf/arch/x86/util/unwind-libdw.c b/tools/perf/arch/x86/util/unwind-libdw.c
new file mode 100644
index 000000000000..c4b72176ca83
--- /dev/null
+++ b/tools/perf/arch/x86/util/unwind-libdw.c
@@ -0,0 +1,51 @@
1#include <elfutils/libdwfl.h>
2#include "../../util/unwind-libdw.h"
3#include "../../util/perf_regs.h"
4
5bool libdw__arch_set_initial_registers(Dwfl_Thread *thread, void *arg)
6{
7 struct unwind_info *ui = arg;
8 struct regs_dump *user_regs = &ui->sample->user_regs;
9 Dwarf_Word dwarf_regs[17];
10 unsigned nregs;
11
12#define REG(r) ({ \
13 Dwarf_Word val = 0; \
14 perf_reg_value(&val, user_regs, PERF_REG_X86_##r); \
15 val; \
16})
17
18 if (user_regs->abi == PERF_SAMPLE_REGS_ABI_32) {
19 dwarf_regs[0] = REG(AX);
20 dwarf_regs[1] = REG(CX);
21 dwarf_regs[2] = REG(DX);
22 dwarf_regs[3] = REG(BX);
23 dwarf_regs[4] = REG(SP);
24 dwarf_regs[5] = REG(BP);
25 dwarf_regs[6] = REG(SI);
26 dwarf_regs[7] = REG(DI);
27 dwarf_regs[8] = REG(IP);
28 nregs = 9;
29 } else {
30 dwarf_regs[0] = REG(AX);
31 dwarf_regs[1] = REG(DX);
32 dwarf_regs[2] = REG(CX);
33 dwarf_regs[3] = REG(BX);
34 dwarf_regs[4] = REG(SI);
35 dwarf_regs[5] = REG(DI);
36 dwarf_regs[6] = REG(BP);
37 dwarf_regs[7] = REG(SP);
38 dwarf_regs[8] = REG(R8);
39 dwarf_regs[9] = REG(R9);
40 dwarf_regs[10] = REG(R10);
41 dwarf_regs[11] = REG(R11);
42 dwarf_regs[12] = REG(R12);
43 dwarf_regs[13] = REG(R13);
44 dwarf_regs[14] = REG(R14);
45 dwarf_regs[15] = REG(R15);
46 dwarf_regs[16] = REG(IP);
47 nregs = 17;
48 }
49
50 return dwfl_thread_state_registers(thread, 0, nregs, dwarf_regs);
51}
diff --git a/tools/perf/arch/x86/util/unwind.c b/tools/perf/arch/x86/util/unwind-libunwind.c
index 456a88cf5b37..3261f68c6a7c 100644
--- a/tools/perf/arch/x86/util/unwind.c
+++ b/tools/perf/arch/x86/util/unwind-libunwind.c
@@ -5,7 +5,7 @@
5#include "../../util/unwind.h" 5#include "../../util/unwind.h"
6 6
7#ifdef HAVE_ARCH_X86_64_SUPPORT 7#ifdef HAVE_ARCH_X86_64_SUPPORT
8int unwind__arch_reg_id(int regnum) 8int libunwind__arch_reg_id(int regnum)
9{ 9{
10 int id; 10 int id;
11 11
@@ -69,7 +69,7 @@ int unwind__arch_reg_id(int regnum)
69 return id; 69 return id;
70} 70}
71#else 71#else
72int unwind__arch_reg_id(int regnum) 72int libunwind__arch_reg_id(int regnum)
73{ 73{
74 int id; 74 int id;
75 75
diff --git a/tools/perf/bench/bench.h b/tools/perf/bench/bench.h
index 0fdc85269c4d..eba46709b279 100644
--- a/tools/perf/bench/bench.h
+++ b/tools/perf/bench/bench.h
@@ -31,6 +31,9 @@ extern int bench_sched_pipe(int argc, const char **argv, const char *prefix);
31extern int bench_mem_memcpy(int argc, const char **argv, 31extern int bench_mem_memcpy(int argc, const char **argv,
32 const char *prefix __maybe_unused); 32 const char *prefix __maybe_unused);
33extern int bench_mem_memset(int argc, const char **argv, const char *prefix); 33extern int bench_mem_memset(int argc, const char **argv, const char *prefix);
34extern int bench_futex_hash(int argc, const char **argv, const char *prefix);
35extern int bench_futex_wake(int argc, const char **argv, const char *prefix);
36extern int bench_futex_requeue(int argc, const char **argv, const char *prefix);
34 37
35#define BENCH_FORMAT_DEFAULT_STR "default" 38#define BENCH_FORMAT_DEFAULT_STR "default"
36#define BENCH_FORMAT_DEFAULT 0 39#define BENCH_FORMAT_DEFAULT 0
diff --git a/tools/perf/bench/futex-hash.c b/tools/perf/bench/futex-hash.c
new file mode 100644
index 000000000000..a84206e9c4aa
--- /dev/null
+++ b/tools/perf/bench/futex-hash.c
@@ -0,0 +1,212 @@
1/*
2 * Copyright (C) 2013 Davidlohr Bueso <davidlohr@hp.com>
3 *
4 * futex-hash: Stress the hell out of the Linux kernel futex uaddr hashing.
5 *
6 * This program is particularly useful for measuring the kernel's futex hash
7 * table/function implementation. In order for it to make sense, use with as
8 * many threads and futexes as possible.
9 */
10
11#include "../perf.h"
12#include "../util/util.h"
13#include "../util/stat.h"
14#include "../util/parse-options.h"
15#include "../util/header.h"
16#include "bench.h"
17#include "futex.h"
18
19#include <err.h>
20#include <stdlib.h>
21#include <sys/time.h>
22#include <pthread.h>
23
24static unsigned int nthreads = 0;
25static unsigned int nsecs = 10;
26/* amount of futexes per thread */
27static unsigned int nfutexes = 1024;
28static bool fshared = false, done = false, silent = false;
29
30struct timeval start, end, runtime;
31static pthread_mutex_t thread_lock;
32static unsigned int threads_starting;
33static struct stats throughput_stats;
34static pthread_cond_t thread_parent, thread_worker;
35
36struct worker {
37 int tid;
38 u_int32_t *futex;
39 pthread_t thread;
40 unsigned long ops;
41};
42
43static const struct option options[] = {
44 OPT_UINTEGER('t', "threads", &nthreads, "Specify amount of threads"),
45 OPT_UINTEGER('r', "runtime", &nsecs, "Specify runtime (in seconds)"),
46 OPT_UINTEGER('f', "futexes", &nfutexes, "Specify amount of futexes per threads"),
47 OPT_BOOLEAN( 's', "silent", &silent, "Silent mode: do not display data/details"),
48 OPT_BOOLEAN( 'S', "shared", &fshared, "Use shared futexes instead of private ones"),
49 OPT_END()
50};
51
52static const char * const bench_futex_hash_usage[] = {
53 "perf bench futex hash <options>",
54 NULL
55};
56
57static void *workerfn(void *arg)
58{
59 int ret;
60 unsigned int i;
61 struct worker *w = (struct worker *) arg;
62
63 pthread_mutex_lock(&thread_lock);
64 threads_starting--;
65 if (!threads_starting)
66 pthread_cond_signal(&thread_parent);
67 pthread_cond_wait(&thread_worker, &thread_lock);
68 pthread_mutex_unlock(&thread_lock);
69
70 do {
71 for (i = 0; i < nfutexes; i++, w->ops++) {
72 /*
73 * We want the futex calls to fail in order to stress
74 * the hashing of uaddr and not measure other steps,
75 * such as internal waitqueue handling, thus enlarging
76 * the critical region protected by hb->lock.
77 */
78 ret = futex_wait(&w->futex[i], 1234, NULL,
79 fshared ? 0 : FUTEX_PRIVATE_FLAG);
80 if (!silent &&
81 (!ret || errno != EAGAIN || errno != EWOULDBLOCK))
82 warn("Non-expected futex return call");
83 }
84 } while (!done);
85
86 return NULL;
87}
88
89static void toggle_done(int sig __maybe_unused,
90 siginfo_t *info __maybe_unused,
91 void *uc __maybe_unused)
92{
93 /* inform all threads that we're done for the day */
94 done = true;
95 gettimeofday(&end, NULL);
96 timersub(&end, &start, &runtime);
97}
98
99static void print_summary(void)
100{
101 unsigned long avg = avg_stats(&throughput_stats);
102 double stddev = stddev_stats(&throughput_stats);
103
104 printf("%sAveraged %ld operations/sec (+- %.2f%%), total secs = %d\n",
105 !silent ? "\n" : "", avg, rel_stddev_stats(stddev, avg),
106 (int) runtime.tv_sec);
107}
108
109int bench_futex_hash(int argc, const char **argv,
110 const char *prefix __maybe_unused)
111{
112 int ret = 0;
113 cpu_set_t cpu;
114 struct sigaction act;
115 unsigned int i, ncpus;
116 pthread_attr_t thread_attr;
117 struct worker *worker = NULL;
118
119 argc = parse_options(argc, argv, options, bench_futex_hash_usage, 0);
120 if (argc) {
121 usage_with_options(bench_futex_hash_usage, options);
122 exit(EXIT_FAILURE);
123 }
124
125 ncpus = sysconf(_SC_NPROCESSORS_ONLN);
126
127 sigfillset(&act.sa_mask);
128 act.sa_sigaction = toggle_done;
129 sigaction(SIGINT, &act, NULL);
130
131 if (!nthreads) /* default to the number of CPUs */
132 nthreads = ncpus;
133
134 worker = calloc(nthreads, sizeof(*worker));
135 if (!worker)
136 goto errmem;
137
138 printf("Run summary [PID %d]: %d threads, each operating on %d [%s] futexes for %d secs.\n\n",
139 getpid(), nthreads, nfutexes, fshared ? "shared":"private", nsecs);
140
141 init_stats(&throughput_stats);
142 pthread_mutex_init(&thread_lock, NULL);
143 pthread_cond_init(&thread_parent, NULL);
144 pthread_cond_init(&thread_worker, NULL);
145
146 threads_starting = nthreads;
147 pthread_attr_init(&thread_attr);
148 gettimeofday(&start, NULL);
149 for (i = 0; i < nthreads; i++) {
150 worker[i].tid = i;
151 worker[i].futex = calloc(nfutexes, sizeof(*worker[i].futex));
152 if (!worker[i].futex)
153 goto errmem;
154
155 CPU_ZERO(&cpu);
156 CPU_SET(i % ncpus, &cpu);
157
158 ret = pthread_attr_setaffinity_np(&thread_attr, sizeof(cpu_set_t), &cpu);
159 if (ret)
160 err(EXIT_FAILURE, "pthread_attr_setaffinity_np");
161
162 ret = pthread_create(&worker[i].thread, &thread_attr, workerfn,
163 (void *)(struct worker *) &worker[i]);
164 if (ret)
165 err(EXIT_FAILURE, "pthread_create");
166
167 }
168 pthread_attr_destroy(&thread_attr);
169
170 pthread_mutex_lock(&thread_lock);
171 while (threads_starting)
172 pthread_cond_wait(&thread_parent, &thread_lock);
173 pthread_cond_broadcast(&thread_worker);
174 pthread_mutex_unlock(&thread_lock);
175
176 sleep(nsecs);
177 toggle_done(0, NULL, NULL);
178
179 for (i = 0; i < nthreads; i++) {
180 ret = pthread_join(worker[i].thread, NULL);
181 if (ret)
182 err(EXIT_FAILURE, "pthread_join");
183 }
184
185 /* cleanup & report results */
186 pthread_cond_destroy(&thread_parent);
187 pthread_cond_destroy(&thread_worker);
188 pthread_mutex_destroy(&thread_lock);
189
190 for (i = 0; i < nthreads; i++) {
191 unsigned long t = worker[i].ops/runtime.tv_sec;
192 update_stats(&throughput_stats, t);
193 if (!silent) {
194 if (nfutexes == 1)
195 printf("[thread %2d] futex: %p [ %ld ops/sec ]\n",
196 worker[i].tid, &worker[i].futex[0], t);
197 else
198 printf("[thread %2d] futexes: %p ... %p [ %ld ops/sec ]\n",
199 worker[i].tid, &worker[i].futex[0],
200 &worker[i].futex[nfutexes-1], t);
201 }
202
203 free(worker[i].futex);
204 }
205
206 print_summary();
207
208 free(worker);
209 return ret;
210errmem:
211 err(EXIT_FAILURE, "calloc");
212}
diff --git a/tools/perf/bench/futex-requeue.c b/tools/perf/bench/futex-requeue.c
new file mode 100644
index 000000000000..a16255876f1d
--- /dev/null
+++ b/tools/perf/bench/futex-requeue.c
@@ -0,0 +1,211 @@
1/*
2 * Copyright (C) 2013 Davidlohr Bueso <davidlohr@hp.com>
3 *
4 * futex-requeue: Block a bunch of threads on futex1 and requeue them
5 * on futex2, N at a time.
6 *
7 * This program is particularly useful to measure the latency of nthread
8 * requeues without waking up any tasks -- thus mimicking a regular futex_wait.
9 */
10
11#include "../perf.h"
12#include "../util/util.h"
13#include "../util/stat.h"
14#include "../util/parse-options.h"
15#include "../util/header.h"
16#include "bench.h"
17#include "futex.h"
18
19#include <err.h>
20#include <stdlib.h>
21#include <sys/time.h>
22#include <pthread.h>
23
24static u_int32_t futex1 = 0, futex2 = 0;
25
26/*
27 * How many tasks to requeue at a time.
28 * Default to 1 in order to make the kernel work more.
29 */
30static unsigned int nrequeue = 1;
31
32/*
33 * There can be significant variance from run to run,
34 * the more repeats, the more exact the overall avg and
35 * the better idea of the futex latency.
36 */
37static unsigned int repeat = 10;
38
39static pthread_t *worker;
40static bool done = 0, silent = 0;
41static pthread_mutex_t thread_lock;
42static pthread_cond_t thread_parent, thread_worker;
43static struct stats requeuetime_stats, requeued_stats;
44static unsigned int ncpus, threads_starting, nthreads = 0;
45
46static const struct option options[] = {
47 OPT_UINTEGER('t', "threads", &nthreads, "Specify amount of threads"),
48 OPT_UINTEGER('q', "nrequeue", &nrequeue, "Specify amount of threads to requeue at once"),
49 OPT_UINTEGER('r', "repeat", &repeat, "Specify amount of times to repeat the run"),
50 OPT_BOOLEAN( 's', "silent", &silent, "Silent mode: do not display data/details"),
51 OPT_END()
52};
53
54static const char * const bench_futex_requeue_usage[] = {
55 "perf bench futex requeue <options>",
56 NULL
57};
58
59static void print_summary(void)
60{
61 double requeuetime_avg = avg_stats(&requeuetime_stats);
62 double requeuetime_stddev = stddev_stats(&requeuetime_stats);
63 unsigned int requeued_avg = avg_stats(&requeued_stats);
64
65 printf("Requeued %d of %d threads in %.4f ms (+-%.2f%%)\n",
66 requeued_avg,
67 nthreads,
68 requeuetime_avg/1e3,
69 rel_stddev_stats(requeuetime_stddev, requeuetime_avg));
70}
71
72static void *workerfn(void *arg __maybe_unused)
73{
74 pthread_mutex_lock(&thread_lock);
75 threads_starting--;
76 if (!threads_starting)
77 pthread_cond_signal(&thread_parent);
78 pthread_cond_wait(&thread_worker, &thread_lock);
79 pthread_mutex_unlock(&thread_lock);
80
81 futex_wait(&futex1, 0, NULL, FUTEX_PRIVATE_FLAG);
82 return NULL;
83}
84
85static void block_threads(pthread_t *w,
86 pthread_attr_t thread_attr)
87{
88 cpu_set_t cpu;
89 unsigned int i;
90
91 threads_starting = nthreads;
92
93 /* create and block all threads */
94 for (i = 0; i < nthreads; i++) {
95 CPU_ZERO(&cpu);
96 CPU_SET(i % ncpus, &cpu);
97
98 if (pthread_attr_setaffinity_np(&thread_attr, sizeof(cpu_set_t), &cpu))
99 err(EXIT_FAILURE, "pthread_attr_setaffinity_np");
100
101 if (pthread_create(&w[i], &thread_attr, workerfn, NULL))
102 err(EXIT_FAILURE, "pthread_create");
103 }
104}
105
106static void toggle_done(int sig __maybe_unused,
107 siginfo_t *info __maybe_unused,
108 void *uc __maybe_unused)
109{
110 done = true;
111}
112
113int bench_futex_requeue(int argc, const char **argv,
114 const char *prefix __maybe_unused)
115{
116 int ret = 0;
117 unsigned int i, j;
118 struct sigaction act;
119 pthread_attr_t thread_attr;
120
121 argc = parse_options(argc, argv, options, bench_futex_requeue_usage, 0);
122 if (argc)
123 goto err;
124
125 ncpus = sysconf(_SC_NPROCESSORS_ONLN);
126
127 sigfillset(&act.sa_mask);
128 act.sa_sigaction = toggle_done;
129 sigaction(SIGINT, &act, NULL);
130
131 if (!nthreads)
132 nthreads = ncpus;
133
134 worker = calloc(nthreads, sizeof(*worker));
135 if (!worker)
136 err(EXIT_FAILURE, "calloc");
137
138 printf("Run summary [PID %d]: Requeuing %d threads (from %p to %p), "
139 "%d at a time.\n\n",
140 getpid(), nthreads, &futex1, &futex2, nrequeue);
141
142 init_stats(&requeued_stats);
143 init_stats(&requeuetime_stats);
144 pthread_attr_init(&thread_attr);
145 pthread_mutex_init(&thread_lock, NULL);
146 pthread_cond_init(&thread_parent, NULL);
147 pthread_cond_init(&thread_worker, NULL);
148
149 for (j = 0; j < repeat && !done; j++) {
150 unsigned int nrequeued = 0;
151 struct timeval start, end, runtime;
152
153 /* create, launch & block all threads */
154 block_threads(worker, thread_attr);
155
156 /* make sure all threads are already blocked */
157 pthread_mutex_lock(&thread_lock);
158 while (threads_starting)
159 pthread_cond_wait(&thread_parent, &thread_lock);
160 pthread_cond_broadcast(&thread_worker);
161 pthread_mutex_unlock(&thread_lock);
162
163 usleep(100000);
164
165 /* Ok, all threads are patiently blocked, start requeueing */
166 gettimeofday(&start, NULL);
167 for (nrequeued = 0; nrequeued < nthreads; nrequeued += nrequeue)
168 /*
169 * Do not wakeup any tasks blocked on futex1, allowing
170 * us to really measure futex_wait functionality.
171 */
172 futex_cmp_requeue(&futex1, 0, &futex2, 0, nrequeue,
173 FUTEX_PRIVATE_FLAG);
174 gettimeofday(&end, NULL);
175 timersub(&end, &start, &runtime);
176
177 update_stats(&requeued_stats, nrequeued);
178 update_stats(&requeuetime_stats, runtime.tv_usec);
179
180 if (!silent) {
181 printf("[Run %d]: Requeued %d of %d threads in %.4f ms\n",
182 j + 1, nrequeued, nthreads, runtime.tv_usec/1e3);
183 }
184
185 /* everybody should be blocked on futex2, wake'em up */
186 nrequeued = futex_wake(&futex2, nthreads, FUTEX_PRIVATE_FLAG);
187 if (nthreads != nrequeued)
188 warnx("couldn't wakeup all tasks (%d/%d)", nrequeued, nthreads);
189
190 for (i = 0; i < nthreads; i++) {
191 ret = pthread_join(worker[i], NULL);
192 if (ret)
193 err(EXIT_FAILURE, "pthread_join");
194 }
195
196 }
197
198 /* cleanup & report results */
199 pthread_cond_destroy(&thread_parent);
200 pthread_cond_destroy(&thread_worker);
201 pthread_mutex_destroy(&thread_lock);
202 pthread_attr_destroy(&thread_attr);
203
204 print_summary();
205
206 free(worker);
207 return ret;
208err:
209 usage_with_options(bench_futex_requeue_usage, options);
210 exit(EXIT_FAILURE);
211}
diff --git a/tools/perf/bench/futex-wake.c b/tools/perf/bench/futex-wake.c
new file mode 100644
index 000000000000..d096169b161e
--- /dev/null
+++ b/tools/perf/bench/futex-wake.c
@@ -0,0 +1,201 @@
1/*
2 * Copyright (C) 2013 Davidlohr Bueso <davidlohr@hp.com>
3 *
4 * futex-wake: Block a bunch of threads on a futex and wake'em up, N at a time.
5 *
6 * This program is particularly useful to measure the latency of nthread wakeups
7 * in non-error situations: all waiters are queued and all wake calls wakeup
8 * one or more tasks, and thus the waitqueue is never empty.
9 */
10
11#include "../perf.h"
12#include "../util/util.h"
13#include "../util/stat.h"
14#include "../util/parse-options.h"
15#include "../util/header.h"
16#include "bench.h"
17#include "futex.h"
18
19#include <err.h>
20#include <stdlib.h>
21#include <sys/time.h>
22#include <pthread.h>
23
24/* all threads will block on the same futex */
25static u_int32_t futex1 = 0;
26
27/*
28 * How many wakeups to do at a time.
29 * Default to 1 in order to make the kernel work more.
30 */
31static unsigned int nwakes = 1;
32
33/*
34 * There can be significant variance from run to run,
35 * the more repeats, the more exact the overall avg and
36 * the better idea of the futex latency.
37 */
38static unsigned int repeat = 10;
39
40pthread_t *worker;
41static bool done = 0, silent = 0;
42static pthread_mutex_t thread_lock;
43static pthread_cond_t thread_parent, thread_worker;
44static struct stats waketime_stats, wakeup_stats;
45static unsigned int ncpus, threads_starting, nthreads = 0;
46
47static const struct option options[] = {
48 OPT_UINTEGER('t', "threads", &nthreads, "Specify amount of threads"),
49 OPT_UINTEGER('w', "nwakes", &nwakes, "Specify amount of threads to wake at once"),
50 OPT_UINTEGER('r', "repeat", &repeat, "Specify amount of times to repeat the run"),
51 OPT_BOOLEAN( 's', "silent", &silent, "Silent mode: do not display data/details"),
52 OPT_END()
53};
54
55static const char * const bench_futex_wake_usage[] = {
56 "perf bench futex wake <options>",
57 NULL
58};
59
60static void *workerfn(void *arg __maybe_unused)
61{
62 pthread_mutex_lock(&thread_lock);
63 threads_starting--;
64 if (!threads_starting)
65 pthread_cond_signal(&thread_parent);
66 pthread_cond_wait(&thread_worker, &thread_lock);
67 pthread_mutex_unlock(&thread_lock);
68
69 futex_wait(&futex1, 0, NULL, FUTEX_PRIVATE_FLAG);
70 return NULL;
71}
72
73static void print_summary(void)
74{
75 double waketime_avg = avg_stats(&waketime_stats);
76 double waketime_stddev = stddev_stats(&waketime_stats);
77 unsigned int wakeup_avg = avg_stats(&wakeup_stats);
78
79 printf("Wokeup %d of %d threads in %.4f ms (+-%.2f%%)\n",
80 wakeup_avg,
81 nthreads,
82 waketime_avg/1e3,
83 rel_stddev_stats(waketime_stddev, waketime_avg));
84}
85
86static void block_threads(pthread_t *w,
87 pthread_attr_t thread_attr)
88{
89 cpu_set_t cpu;
90 unsigned int i;
91
92 threads_starting = nthreads;
93
94 /* create and block all threads */
95 for (i = 0; i < nthreads; i++) {
96 CPU_ZERO(&cpu);
97 CPU_SET(i % ncpus, &cpu);
98
99 if (pthread_attr_setaffinity_np(&thread_attr, sizeof(cpu_set_t), &cpu))
100 err(EXIT_FAILURE, "pthread_attr_setaffinity_np");
101
102 if (pthread_create(&w[i], &thread_attr, workerfn, NULL))
103 err(EXIT_FAILURE, "pthread_create");
104 }
105}
106
107static void toggle_done(int sig __maybe_unused,
108 siginfo_t *info __maybe_unused,
109 void *uc __maybe_unused)
110{
111 done = true;
112}
113
114int bench_futex_wake(int argc, const char **argv,
115 const char *prefix __maybe_unused)
116{
117 int ret = 0;
118 unsigned int i, j;
119 struct sigaction act;
120 pthread_attr_t thread_attr;
121
122 argc = parse_options(argc, argv, options, bench_futex_wake_usage, 0);
123 if (argc) {
124 usage_with_options(bench_futex_wake_usage, options);
125 exit(EXIT_FAILURE);
126 }
127
128 ncpus = sysconf(_SC_NPROCESSORS_ONLN);
129
130 sigfillset(&act.sa_mask);
131 act.sa_sigaction = toggle_done;
132 sigaction(SIGINT, &act, NULL);
133
134 if (!nthreads)
135 nthreads = ncpus;
136
137 worker = calloc(nthreads, sizeof(*worker));
138 if (!worker)
139 err(EXIT_FAILURE, "calloc");
140
141 printf("Run summary [PID %d]: blocking on %d threads (at futex %p), "
142 "waking up %d at a time.\n\n",
143 getpid(), nthreads, &futex1, nwakes);
144
145 init_stats(&wakeup_stats);
146 init_stats(&waketime_stats);
147 pthread_attr_init(&thread_attr);
148 pthread_mutex_init(&thread_lock, NULL);
149 pthread_cond_init(&thread_parent, NULL);
150 pthread_cond_init(&thread_worker, NULL);
151
152 for (j = 0; j < repeat && !done; j++) {
153 unsigned int nwoken = 0;
154 struct timeval start, end, runtime;
155
156 /* create, launch & block all threads */
157 block_threads(worker, thread_attr);
158
159 /* make sure all threads are already blocked */
160 pthread_mutex_lock(&thread_lock);
161 while (threads_starting)
162 pthread_cond_wait(&thread_parent, &thread_lock);
163 pthread_cond_broadcast(&thread_worker);
164 pthread_mutex_unlock(&thread_lock);
165
166 usleep(100000);
167
168 /* Ok, all threads are patiently blocked, start waking folks up */
169 gettimeofday(&start, NULL);
170 while (nwoken != nthreads)
171 nwoken += futex_wake(&futex1, nwakes, FUTEX_PRIVATE_FLAG);
172 gettimeofday(&end, NULL);
173 timersub(&end, &start, &runtime);
174
175 update_stats(&wakeup_stats, nwoken);
176 update_stats(&waketime_stats, runtime.tv_usec);
177
178 if (!silent) {
179 printf("[Run %d]: Wokeup %d of %d threads in %.4f ms\n",
180 j + 1, nwoken, nthreads, runtime.tv_usec/1e3);
181 }
182
183 for (i = 0; i < nthreads; i++) {
184 ret = pthread_join(worker[i], NULL);
185 if (ret)
186 err(EXIT_FAILURE, "pthread_join");
187 }
188
189 }
190
191 /* cleanup & report results */
192 pthread_cond_destroy(&thread_parent);
193 pthread_cond_destroy(&thread_worker);
194 pthread_mutex_destroy(&thread_lock);
195 pthread_attr_destroy(&thread_attr);
196
197 print_summary();
198
199 free(worker);
200 return ret;
201}
diff --git a/tools/perf/bench/futex.h b/tools/perf/bench/futex.h
new file mode 100644
index 000000000000..71f2844cf97f
--- /dev/null
+++ b/tools/perf/bench/futex.h
@@ -0,0 +1,71 @@
1/*
2 * Glibc independent futex library for testing kernel functionality.
3 * Shamelessly stolen from Darren Hart <dvhltc@us.ibm.com>
4 * http://git.kernel.org/cgit/linux/kernel/git/dvhart/futextest.git/
5 */
6
7#ifndef _FUTEX_H
8#define _FUTEX_H
9
10#include <unistd.h>
11#include <sys/syscall.h>
12#include <sys/types.h>
13#include <linux/futex.h>
14
15/**
16 * futex() - SYS_futex syscall wrapper
17 * @uaddr: address of first futex
18 * @op: futex op code
19 * @val: typically expected value of uaddr, but varies by op
20 * @timeout: typically an absolute struct timespec (except where noted
21 * otherwise). Overloaded by some ops
22 * @uaddr2: address of second futex for some ops\
23 * @val3: varies by op
24 * @opflags: flags to be bitwise OR'd with op, such as FUTEX_PRIVATE_FLAG
25 *
26 * futex() is used by all the following futex op wrappers. It can also be
27 * used for misuse and abuse testing. Generally, the specific op wrappers
28 * should be used instead. It is a macro instead of an static inline function as
29 * some of the types over overloaded (timeout is used for nr_requeue for
30 * example).
31 *
32 * These argument descriptions are the defaults for all
33 * like-named arguments in the following wrappers except where noted below.
34 */
35#define futex(uaddr, op, val, timeout, uaddr2, val3, opflags) \
36 syscall(SYS_futex, uaddr, op | opflags, val, timeout, uaddr2, val3)
37
38/**
39 * futex_wait() - block on uaddr with optional timeout
40 * @timeout: relative timeout
41 */
42static inline int
43futex_wait(u_int32_t *uaddr, u_int32_t val, struct timespec *timeout, int opflags)
44{
45 return futex(uaddr, FUTEX_WAIT, val, timeout, NULL, 0, opflags);
46}
47
48/**
49 * futex_wake() - wake one or more tasks blocked on uaddr
50 * @nr_wake: wake up to this many tasks
51 */
52static inline int
53futex_wake(u_int32_t *uaddr, int nr_wake, int opflags)
54{
55 return futex(uaddr, FUTEX_WAKE, nr_wake, NULL, NULL, 0, opflags);
56}
57
58/**
59* futex_cmp_requeue() - requeue tasks from uaddr to uaddr2
60* @nr_wake: wake up to this many tasks
61* @nr_requeue: requeue up to this many tasks
62*/
63static inline int
64futex_cmp_requeue(u_int32_t *uaddr, u_int32_t val, u_int32_t *uaddr2, int nr_wake,
65 int nr_requeue, int opflags)
66{
67 return futex(uaddr, FUTEX_CMP_REQUEUE, nr_wake, nr_requeue, uaddr2,
68 val, opflags);
69}
70
71#endif /* _FUTEX_H */
diff --git a/tools/perf/builtin-bench.c b/tools/perf/builtin-bench.c
index 8a987d252780..1e6e77710545 100644
--- a/tools/perf/builtin-bench.c
+++ b/tools/perf/builtin-bench.c
@@ -12,6 +12,7 @@
12 * sched ... scheduler and IPC performance 12 * sched ... scheduler and IPC performance
13 * mem ... memory access performance 13 * mem ... memory access performance
14 * numa ... NUMA scheduling and MM performance 14 * numa ... NUMA scheduling and MM performance
15 * futex ... Futex performance
15 */ 16 */
16#include "perf.h" 17#include "perf.h"
17#include "util/util.h" 18#include "util/util.h"
@@ -54,6 +55,14 @@ static struct bench mem_benchmarks[] = {
54 { NULL, NULL, NULL } 55 { NULL, NULL, NULL }
55}; 56};
56 57
58static struct bench futex_benchmarks[] = {
59 { "hash", "Benchmark for futex hash table", bench_futex_hash },
60 { "wake", "Benchmark for futex wake calls", bench_futex_wake },
61 { "requeue", "Benchmark for futex requeue calls", bench_futex_requeue },
62 { "all", "Test all futex benchmarks", NULL },
63 { NULL, NULL, NULL }
64};
65
57struct collection { 66struct collection {
58 const char *name; 67 const char *name;
59 const char *summary; 68 const char *summary;
@@ -61,11 +70,12 @@ struct collection {
61}; 70};
62 71
63static struct collection collections[] = { 72static struct collection collections[] = {
64 { "sched", "Scheduler and IPC benchmarks", sched_benchmarks }, 73 { "sched", "Scheduler and IPC benchmarks", sched_benchmarks },
65 { "mem", "Memory access benchmarks", mem_benchmarks }, 74 { "mem", "Memory access benchmarks", mem_benchmarks },
66#ifdef HAVE_LIBNUMA_SUPPORT 75#ifdef HAVE_LIBNUMA_SUPPORT
67 { "numa", "NUMA scheduling and MM benchmarks", numa_benchmarks }, 76 { "numa", "NUMA scheduling and MM benchmarks", numa_benchmarks },
68#endif 77#endif
78 {"futex", "Futex stressing benchmarks", futex_benchmarks },
69 { "all", "All benchmarks", NULL }, 79 { "all", "All benchmarks", NULL },
70 { NULL, NULL, NULL } 80 { NULL, NULL, NULL }
71}; 81};
diff --git a/tools/perf/builtin-diff.c b/tools/perf/builtin-diff.c
index a77e31246c00..204fffe22532 100644
--- a/tools/perf/builtin-diff.c
+++ b/tools/perf/builtin-diff.c
@@ -952,8 +952,8 @@ static int hpp__entry_global(struct perf_hpp_fmt *_fmt, struct perf_hpp *hpp,
952 dfmt->header_width, buf); 952 dfmt->header_width, buf);
953} 953}
954 954
955static int hpp__header(struct perf_hpp_fmt *fmt, 955static int hpp__header(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
956 struct perf_hpp *hpp) 956 struct perf_evsel *evsel __maybe_unused)
957{ 957{
958 struct diff_hpp_fmt *dfmt = 958 struct diff_hpp_fmt *dfmt =
959 container_of(fmt, struct diff_hpp_fmt, fmt); 959 container_of(fmt, struct diff_hpp_fmt, fmt);
@@ -963,7 +963,8 @@ static int hpp__header(struct perf_hpp_fmt *fmt,
963} 963}
964 964
965static int hpp__width(struct perf_hpp_fmt *fmt, 965static int hpp__width(struct perf_hpp_fmt *fmt,
966 struct perf_hpp *hpp __maybe_unused) 966 struct perf_hpp *hpp __maybe_unused,
967 struct perf_evsel *evsel __maybe_unused)
967{ 968{
968 struct diff_hpp_fmt *dfmt = 969 struct diff_hpp_fmt *dfmt =
969 container_of(fmt, struct diff_hpp_fmt, fmt); 970 container_of(fmt, struct diff_hpp_fmt, fmt);
diff --git a/tools/perf/builtin-inject.c b/tools/perf/builtin-inject.c
index b3466018bbd7..3a7387551369 100644
--- a/tools/perf/builtin-inject.c
+++ b/tools/perf/builtin-inject.c
@@ -312,7 +312,6 @@ found:
312 sample_sw.period = sample->period; 312 sample_sw.period = sample->period;
313 sample_sw.time = sample->time; 313 sample_sw.time = sample->time;
314 perf_event__synthesize_sample(event_sw, evsel->attr.sample_type, 314 perf_event__synthesize_sample(event_sw, evsel->attr.sample_type,
315 evsel->attr.sample_regs_user,
316 evsel->attr.read_format, &sample_sw, 315 evsel->attr.read_format, &sample_sw,
317 false); 316 false);
318 build_id__mark_dso_hit(tool, event_sw, &sample_sw, evsel, machine); 317 build_id__mark_dso_hit(tool, event_sw, &sample_sw, evsel, machine);
diff --git a/tools/perf/builtin-kvm.c b/tools/perf/builtin-kvm.c
index a7350519c63f..21c164b8f9db 100644
--- a/tools/perf/builtin-kvm.c
+++ b/tools/perf/builtin-kvm.c
@@ -1691,17 +1691,15 @@ int cmd_kvm(int argc, const char **argv, const char *prefix __maybe_unused)
1691 OPT_END() 1691 OPT_END()
1692 }; 1692 };
1693 1693
1694 1694 const char *const kvm_subcommands[] = { "top", "record", "report", "diff",
1695 const char * const kvm_usage[] = { 1695 "buildid-list", "stat", NULL };
1696 "perf kvm [<options>] {top|record|report|diff|buildid-list|stat}", 1696 const char *kvm_usage[] = { NULL, NULL };
1697 NULL
1698 };
1699 1697
1700 perf_host = 0; 1698 perf_host = 0;
1701 perf_guest = 1; 1699 perf_guest = 1;
1702 1700
1703 argc = parse_options(argc, argv, kvm_options, kvm_usage, 1701 argc = parse_options_subcommand(argc, argv, kvm_options, kvm_subcommands, kvm_usage,
1704 PARSE_OPT_STOP_AT_NON_OPTION); 1702 PARSE_OPT_STOP_AT_NON_OPTION);
1705 if (!argc) 1703 if (!argc)
1706 usage_with_options(kvm_usage, kvm_options); 1704 usage_with_options(kvm_usage, kvm_options);
1707 1705
diff --git a/tools/perf/builtin-probe.c b/tools/perf/builtin-probe.c
index 78948882e3de..cdcd4eb3a57d 100644
--- a/tools/perf/builtin-probe.c
+++ b/tools/perf/builtin-probe.c
@@ -268,9 +268,9 @@ static int opt_set_filter(const struct option *opt __maybe_unused,
268 return 0; 268 return 0;
269} 269}
270 270
271static void init_params(void) 271static int init_params(void)
272{ 272{
273 line_range__init(&params.line_range); 273 return line_range__init(&params.line_range);
274} 274}
275 275
276static void cleanup_params(void) 276static void cleanup_params(void)
@@ -515,9 +515,11 @@ int cmd_probe(int argc, const char **argv, const char *prefix)
515{ 515{
516 int ret; 516 int ret;
517 517
518 init_params(); 518 ret = init_params();
519 ret = __cmd_probe(argc, argv, prefix); 519 if (!ret) {
520 cleanup_params(); 520 ret = __cmd_probe(argc, argv, prefix);
521 cleanup_params();
522 }
521 523
522 return ret; 524 return ret;
523} 525}
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index af47531b82ec..eb524f91bffe 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -649,7 +649,7 @@ error:
649 return ret; 649 return ret;
650} 650}
651 651
652#ifdef HAVE_LIBUNWIND_SUPPORT 652#ifdef HAVE_DWARF_UNWIND_SUPPORT
653static int get_stack_size(char *str, unsigned long *_size) 653static int get_stack_size(char *str, unsigned long *_size)
654{ 654{
655 char *endptr; 655 char *endptr;
@@ -675,7 +675,7 @@ static int get_stack_size(char *str, unsigned long *_size)
675 max_size, str); 675 max_size, str);
676 return -1; 676 return -1;
677} 677}
678#endif /* HAVE_LIBUNWIND_SUPPORT */ 678#endif /* HAVE_DWARF_UNWIND_SUPPORT */
679 679
680int record_parse_callchain(const char *arg, struct record_opts *opts) 680int record_parse_callchain(const char *arg, struct record_opts *opts)
681{ 681{
@@ -704,7 +704,7 @@ int record_parse_callchain(const char *arg, struct record_opts *opts)
704 "needed for -g fp\n"); 704 "needed for -g fp\n");
705 break; 705 break;
706 706
707#ifdef HAVE_LIBUNWIND_SUPPORT 707#ifdef HAVE_DWARF_UNWIND_SUPPORT
708 /* Dwarf style */ 708 /* Dwarf style */
709 } else if (!strncmp(name, "dwarf", sizeof("dwarf"))) { 709 } else if (!strncmp(name, "dwarf", sizeof("dwarf"))) {
710 const unsigned long default_stack_dump_size = 8192; 710 const unsigned long default_stack_dump_size = 8192;
@@ -720,7 +720,7 @@ int record_parse_callchain(const char *arg, struct record_opts *opts)
720 ret = get_stack_size(tok, &size); 720 ret = get_stack_size(tok, &size);
721 opts->stack_dump_size = size; 721 opts->stack_dump_size = size;
722 } 722 }
723#endif /* HAVE_LIBUNWIND_SUPPORT */ 723#endif /* HAVE_DWARF_UNWIND_SUPPORT */
724 } else { 724 } else {
725 pr_err("callchain: Unknown --call-graph option " 725 pr_err("callchain: Unknown --call-graph option "
726 "value: %s\n", arg); 726 "value: %s\n", arg);
@@ -735,7 +735,9 @@ int record_parse_callchain(const char *arg, struct record_opts *opts)
735 735
736static void callchain_debug(struct record_opts *opts) 736static void callchain_debug(struct record_opts *opts)
737{ 737{
738 pr_debug("callchain: type %d\n", opts->call_graph); 738 static const char *str[CALLCHAIN_MAX] = { "NONE", "FP", "DWARF" };
739
740 pr_debug("callchain: type %s\n", str[opts->call_graph]);
739 741
740 if (opts->call_graph == CALLCHAIN_DWARF) 742 if (opts->call_graph == CALLCHAIN_DWARF)
741 pr_debug("callchain: stack dump size %d\n", 743 pr_debug("callchain: stack dump size %d\n",
@@ -749,6 +751,8 @@ int record_parse_callchain_opt(const struct option *opt,
749 struct record_opts *opts = opt->value; 751 struct record_opts *opts = opt->value;
750 int ret; 752 int ret;
751 753
754 opts->call_graph_enabled = !unset;
755
752 /* --no-call-graph */ 756 /* --no-call-graph */
753 if (unset) { 757 if (unset) {
754 opts->call_graph = CALLCHAIN_NONE; 758 opts->call_graph = CALLCHAIN_NONE;
@@ -769,6 +773,8 @@ int record_callchain_opt(const struct option *opt,
769{ 773{
770 struct record_opts *opts = opt->value; 774 struct record_opts *opts = opt->value;
771 775
776 opts->call_graph_enabled = !unset;
777
772 if (opts->call_graph == CALLCHAIN_NONE) 778 if (opts->call_graph == CALLCHAIN_NONE)
773 opts->call_graph = CALLCHAIN_FP; 779 opts->call_graph = CALLCHAIN_FP;
774 780
@@ -776,6 +782,16 @@ int record_callchain_opt(const struct option *opt,
776 return 0; 782 return 0;
777} 783}
778 784
785static int perf_record_config(const char *var, const char *value, void *cb)
786{
787 struct record *rec = cb;
788
789 if (!strcmp(var, "record.call-graph"))
790 return record_parse_callchain(value, &rec->opts);
791
792 return perf_default_config(var, value, cb);
793}
794
779static const char * const record_usage[] = { 795static const char * const record_usage[] = {
780 "perf record [<options>] [<command>]", 796 "perf record [<options>] [<command>]",
781 "perf record [<options>] -- <command> [<options>]", 797 "perf record [<options>] -- <command> [<options>]",
@@ -807,7 +823,7 @@ static struct record record = {
807 823
808#define CALLCHAIN_HELP "setup and enables call-graph (stack chain/backtrace) recording: " 824#define CALLCHAIN_HELP "setup and enables call-graph (stack chain/backtrace) recording: "
809 825
810#ifdef HAVE_LIBUNWIND_SUPPORT 826#ifdef HAVE_DWARF_UNWIND_SUPPORT
811const char record_callchain_help[] = CALLCHAIN_HELP "fp dwarf"; 827const char record_callchain_help[] = CALLCHAIN_HELP "fp dwarf";
812#else 828#else
813const char record_callchain_help[] = CALLCHAIN_HELP "fp"; 829const char record_callchain_help[] = CALLCHAIN_HELP "fp";
@@ -907,6 +923,8 @@ int cmd_record(int argc, const char **argv, const char *prefix __maybe_unused)
907 if (rec->evlist == NULL) 923 if (rec->evlist == NULL)
908 return -ENOMEM; 924 return -ENOMEM;
909 925
926 perf_config(perf_record_config, rec);
927
910 argc = parse_options(argc, argv, record_options, record_usage, 928 argc = parse_options(argc, argv, record_options, record_usage,
911 PARSE_OPT_STOP_AT_NON_OPTION); 929 PARSE_OPT_STOP_AT_NON_OPTION);
912 if (!argc && target__none(&rec->opts.target)) 930 if (!argc && target__none(&rec->opts.target))
diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index 02f985f3a396..c8f21137dfd8 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -75,13 +75,10 @@ static int report__config(const char *var, const char *value, void *cb)
75 return perf_default_config(var, value, cb); 75 return perf_default_config(var, value, cb);
76} 76}
77 77
78static int report__add_mem_hist_entry(struct perf_tool *tool, struct addr_location *al, 78static int report__add_mem_hist_entry(struct report *rep, struct addr_location *al,
79 struct perf_sample *sample, struct perf_evsel *evsel, 79 struct perf_sample *sample, struct perf_evsel *evsel)
80 union perf_event *event)
81{ 80{
82 struct report *rep = container_of(tool, struct report, tool);
83 struct symbol *parent = NULL; 81 struct symbol *parent = NULL;
84 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
85 struct hist_entry *he; 82 struct hist_entry *he;
86 struct mem_info *mi, *mx; 83 struct mem_info *mi, *mx;
87 uint64_t cost; 84 uint64_t cost;
@@ -90,7 +87,7 @@ static int report__add_mem_hist_entry(struct perf_tool *tool, struct addr_locati
90 if (err) 87 if (err)
91 return err; 88 return err;
92 89
93 mi = machine__resolve_mem(al->machine, al->thread, sample, cpumode); 90 mi = sample__resolve_mem(sample, al);
94 if (!mi) 91 if (!mi)
95 return -ENOMEM; 92 return -ENOMEM;
96 93
@@ -131,10 +128,9 @@ out:
131 return err; 128 return err;
132} 129}
133 130
134static int report__add_branch_hist_entry(struct perf_tool *tool, struct addr_location *al, 131static int report__add_branch_hist_entry(struct report *rep, struct addr_location *al,
135 struct perf_sample *sample, struct perf_evsel *evsel) 132 struct perf_sample *sample, struct perf_evsel *evsel)
136{ 133{
137 struct report *rep = container_of(tool, struct report, tool);
138 struct symbol *parent = NULL; 134 struct symbol *parent = NULL;
139 unsigned i; 135 unsigned i;
140 struct hist_entry *he; 136 struct hist_entry *he;
@@ -144,8 +140,7 @@ static int report__add_branch_hist_entry(struct perf_tool *tool, struct addr_loc
144 if (err) 140 if (err)
145 return err; 141 return err;
146 142
147 bi = machine__resolve_bstack(al->machine, al->thread, 143 bi = sample__resolve_bstack(sample, al);
148 sample->branch_stack);
149 if (!bi) 144 if (!bi)
150 return -ENOMEM; 145 return -ENOMEM;
151 146
@@ -190,10 +185,9 @@ out:
190 return err; 185 return err;
191} 186}
192 187
193static int report__add_hist_entry(struct perf_tool *tool, struct perf_evsel *evsel, 188static int report__add_hist_entry(struct report *rep, struct perf_evsel *evsel,
194 struct addr_location *al, struct perf_sample *sample) 189 struct addr_location *al, struct perf_sample *sample)
195{ 190{
196 struct report *rep = container_of(tool, struct report, tool);
197 struct symbol *parent = NULL; 191 struct symbol *parent = NULL;
198 struct hist_entry *he; 192 struct hist_entry *he;
199 int err = sample__resolve_callchain(sample, &parent, evsel, al, rep->max_stack); 193 int err = sample__resolve_callchain(sample, &parent, evsel, al, rep->max_stack);
@@ -237,25 +231,25 @@ static int process_sample_event(struct perf_tool *tool,
237 return -1; 231 return -1;
238 } 232 }
239 233
240 if (al.filtered || (rep->hide_unresolved && al.sym == NULL)) 234 if (rep->hide_unresolved && al.sym == NULL)
241 return 0; 235 return 0;
242 236
243 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 237 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
244 return 0; 238 return 0;
245 239
246 if (sort__mode == SORT_MODE__BRANCH) { 240 if (sort__mode == SORT_MODE__BRANCH) {
247 ret = report__add_branch_hist_entry(tool, &al, sample, evsel); 241 ret = report__add_branch_hist_entry(rep, &al, sample, evsel);
248 if (ret < 0) 242 if (ret < 0)
249 pr_debug("problem adding lbr entry, skipping event\n"); 243 pr_debug("problem adding lbr entry, skipping event\n");
250 } else if (rep->mem_mode == 1) { 244 } else if (rep->mem_mode == 1) {
251 ret = report__add_mem_hist_entry(tool, &al, sample, evsel, event); 245 ret = report__add_mem_hist_entry(rep, &al, sample, evsel);
252 if (ret < 0) 246 if (ret < 0)
253 pr_debug("problem adding mem entry, skipping event\n"); 247 pr_debug("problem adding mem entry, skipping event\n");
254 } else { 248 } else {
255 if (al.map != NULL) 249 if (al.map != NULL)
256 al.map->dso->hit = 1; 250 al.map->dso->hit = 1;
257 251
258 ret = report__add_hist_entry(tool, evsel, &al, sample); 252 ret = report__add_hist_entry(rep, evsel, &al, sample);
259 if (ret < 0) 253 if (ret < 0)
260 pr_debug("problem incrementing symbol period, skipping event\n"); 254 pr_debug("problem incrementing symbol period, skipping event\n");
261 } 255 }
@@ -934,7 +928,7 @@ repeat:
934 * so don't allocate extra space that won't be used in the stdio 928 * so don't allocate extra space that won't be used in the stdio
935 * implementation. 929 * implementation.
936 */ 930 */
937 if (use_browser == 1 && sort__has_sym) { 931 if (ui__has_annotation()) {
938 symbol_conf.priv_size = sizeof(struct annotation); 932 symbol_conf.priv_size = sizeof(struct annotation);
939 machines__set_symbol_filter(&session->machines, 933 machines__set_symbol_filter(&session->machines,
940 symbol__annotate_init); 934 symbol__annotate_init);
diff --git a/tools/perf/builtin-sched.c b/tools/perf/builtin-sched.c
index 6a76a07b6789..9ac0a495c954 100644
--- a/tools/perf/builtin-sched.c
+++ b/tools/perf/builtin-sched.c
@@ -1124,7 +1124,7 @@ static void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_
1124 1124
1125 avg = work_list->total_lat / work_list->nb_atoms; 1125 avg = work_list->total_lat / work_list->nb_atoms;
1126 1126
1127 printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %9.6f s\n", 1127 printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %13.6f s\n",
1128 (double)work_list->total_runtime / 1e6, 1128 (double)work_list->total_runtime / 1e6,
1129 work_list->nb_atoms, (double)avg / 1e6, 1129 work_list->nb_atoms, (double)avg / 1e6,
1130 (double)work_list->max_lat / 1e6, 1130 (double)work_list->max_lat / 1e6,
@@ -1527,9 +1527,9 @@ static int perf_sched__lat(struct perf_sched *sched)
1527 1527
1528 perf_sched__sort_lat(sched); 1528 perf_sched__sort_lat(sched);
1529 1529
1530 printf("\n ---------------------------------------------------------------------------------------------------------------\n"); 1530 printf("\n -----------------------------------------------------------------------------------------------------------------\n");
1531 printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms | Maximum delay at |\n"); 1531 printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms | Maximum delay at |\n");
1532 printf(" ---------------------------------------------------------------------------------------------------------------\n"); 1532 printf(" -----------------------------------------------------------------------------------------------------------------\n");
1533 1533
1534 next = rb_first(&sched->sorted_atom_root); 1534 next = rb_first(&sched->sorted_atom_root);
1535 1535
@@ -1541,7 +1541,7 @@ static int perf_sched__lat(struct perf_sched *sched)
1541 next = rb_next(next); 1541 next = rb_next(next);
1542 } 1542 }
1543 1543
1544 printf(" -----------------------------------------------------------------------------------------\n"); 1544 printf(" -----------------------------------------------------------------------------------------------------------------\n");
1545 printf(" TOTAL: |%11.3f ms |%9" PRIu64 " |\n", 1545 printf(" TOTAL: |%11.3f ms |%9" PRIu64 " |\n",
1546 (double)sched->all_runtime / 1e6, sched->all_count); 1546 (double)sched->all_runtime / 1e6, sched->all_count);
1547 1547
diff --git a/tools/perf/builtin-timechart.c b/tools/perf/builtin-timechart.c
index 25526d6eae59..74db2568b867 100644
--- a/tools/perf/builtin-timechart.c
+++ b/tools/perf/builtin-timechart.c
@@ -494,7 +494,7 @@ static const char *cat_backtrace(union perf_event *event,
494 continue; 494 continue;
495 } 495 }
496 496
497 tal.filtered = false; 497 tal.filtered = 0;
498 thread__find_addr_location(al.thread, machine, cpumode, 498 thread__find_addr_location(al.thread, machine, cpumode,
499 MAP__FUNCTION, ip, &tal); 499 MAP__FUNCTION, ip, &tal);
500 500
@@ -1238,7 +1238,7 @@ static int timechart__record(struct timechart *tchart, int argc, const char **ar
1238 for (i = 0; i < old_power_args_nr; i++) 1238 for (i = 0; i < old_power_args_nr; i++)
1239 *p++ = strdup(old_power_args[i]); 1239 *p++ = strdup(old_power_args[i]);
1240 1240
1241 for (j = 1; j < (unsigned int)argc; j++) 1241 for (j = 0; j < (unsigned int)argc; j++)
1242 *p++ = argv[j]; 1242 *p++ = argv[j];
1243 1243
1244 return cmd_record(rec_argc, rec_argv, NULL); 1244 return cmd_record(rec_argc, rec_argv, NULL);
diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index 5f989a7d8bc2..65aaa5bbf7ec 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -993,6 +993,16 @@ parse_callchain_opt(const struct option *opt, const char *arg, int unset)
993 return record_parse_callchain_opt(opt, arg, unset); 993 return record_parse_callchain_opt(opt, arg, unset);
994} 994}
995 995
996static int perf_top_config(const char *var, const char *value, void *cb)
997{
998 struct perf_top *top = cb;
999
1000 if (!strcmp(var, "top.call-graph"))
1001 return record_parse_callchain(value, &top->record_opts);
1002
1003 return perf_default_config(var, value, cb);
1004}
1005
996static int 1006static int
997parse_percent_limit(const struct option *opt, const char *arg, 1007parse_percent_limit(const struct option *opt, const char *arg,
998 int unset __maybe_unused) 1008 int unset __maybe_unused)
@@ -1117,6 +1127,8 @@ int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
1117 if (top.evlist == NULL) 1127 if (top.evlist == NULL)
1118 return -ENOMEM; 1128 return -ENOMEM;
1119 1129
1130 perf_config(perf_top_config, &top);
1131
1120 argc = parse_options(argc, argv, options, top_usage, 0); 1132 argc = parse_options(argc, argv, options, top_usage, 0);
1121 if (argc) 1133 if (argc)
1122 usage_with_options(top_usage, options); 1134 usage_with_options(top_usage, options);
diff --git a/tools/perf/config/Makefile b/tools/perf/config/Makefile
index 0331ea2701a3..c23418225c2c 100644
--- a/tools/perf/config/Makefile
+++ b/tools/perf/config/Makefile
@@ -59,6 +59,18 @@ ifeq ($(NO_PERF_REGS),0)
59 CFLAGS += -DHAVE_PERF_REGS_SUPPORT 59 CFLAGS += -DHAVE_PERF_REGS_SUPPORT
60endif 60endif
61 61
62ifndef NO_LIBELF
63 # for linking with debug library, run like:
64 # make DEBUG=1 LIBDW_DIR=/opt/libdw/
65 ifdef LIBDW_DIR
66 LIBDW_CFLAGS := -I$(LIBDW_DIR)/include
67 LIBDW_LDFLAGS := -L$(LIBDW_DIR)/lib
68
69 FEATURE_CHECK_CFLAGS-libdw-dwarf-unwind := $(LIBDW_CFLAGS)
70 FEATURE_CHECK_LDFLAGS-libdw-dwarf-unwind := $(LIBDW_LDFLAGS) -ldw
71 endif
72endif
73
62# include ARCH specific config 74# include ARCH specific config
63-include $(src-perf)/arch/$(ARCH)/Makefile 75-include $(src-perf)/arch/$(ARCH)/Makefile
64 76
@@ -147,7 +159,35 @@ CORE_FEATURE_TESTS = \
147 libunwind \ 159 libunwind \
148 on-exit \ 160 on-exit \
149 stackprotector-all \ 161 stackprotector-all \
150 timerfd 162 timerfd \
163 libdw-dwarf-unwind
164
165LIB_FEATURE_TESTS = \
166 dwarf \
167 glibc \
168 gtk2 \
169 libaudit \
170 libbfd \
171 libelf \
172 libnuma \
173 libperl \
174 libpython \
175 libslang \
176 libunwind \
177 libdw-dwarf-unwind
178
179VF_FEATURE_TESTS = \
180 backtrace \
181 fortify-source \
182 gtk2-infobar \
183 libelf-getphdrnum \
184 libelf-mmap \
185 libpython-version \
186 on-exit \
187 stackprotector-all \
188 timerfd \
189 libunwind-debug-frame \
190 bionic
151 191
152# Set FEATURE_CHECK_(C|LD)FLAGS-all for all CORE_FEATURE_TESTS features. 192# Set FEATURE_CHECK_(C|LD)FLAGS-all for all CORE_FEATURE_TESTS features.
153# If in the future we need per-feature checks/flags for features not 193# If in the future we need per-feature checks/flags for features not
@@ -161,17 +201,6 @@ endef
161$(foreach feat,$(CORE_FEATURE_TESTS),$(call set_test_all_flags,$(feat))) 201$(foreach feat,$(CORE_FEATURE_TESTS),$(call set_test_all_flags,$(feat)))
162 202
163# 203#
164# So here we detect whether test-all was rebuilt, to be able
165# to skip the print-out of the long features list if the file
166# existed before and after it was built:
167#
168ifeq ($(wildcard $(OUTPUT)config/feature-checks/test-all.bin),)
169 test-all-failed := 1
170else
171 test-all-failed := 0
172endif
173
174#
175# Special fast-path for the 'all features are available' case: 204# Special fast-path for the 'all features are available' case:
176# 205#
177$(call feature_check,all,$(MSG)) 206$(call feature_check,all,$(MSG))
@@ -180,15 +209,6 @@ $(call feature_check,all,$(MSG))
180# Just in case the build freshly failed, make sure we print the 209# Just in case the build freshly failed, make sure we print the
181# feature matrix: 210# feature matrix:
182# 211#
183ifeq ($(feature-all), 0)
184 test-all-failed := 1
185endif
186
187ifeq ($(test-all-failed),1)
188 $(info )
189 $(info Auto-detecting system features:)
190endif
191
192ifeq ($(feature-all), 1) 212ifeq ($(feature-all), 1)
193 # 213 #
194 # test-all.c passed - just set all the core feature flags to 1: 214 # test-all.c passed - just set all the core feature flags to 1:
@@ -199,27 +219,6 @@ else
199 $(foreach feat,$(CORE_FEATURE_TESTS),$(call feature_check,$(feat))) 219 $(foreach feat,$(CORE_FEATURE_TESTS),$(call feature_check,$(feat)))
200endif 220endif
201 221
202#
203# Print the result of the feature test:
204#
205feature_print = $(eval $(feature_print_code)) $(info $(MSG))
206
207define feature_print_code
208 ifeq ($(feature-$(1)), 1)
209 MSG = $(shell printf '...%30s: [ \033[32mon\033[m ]' $(1))
210 else
211 MSG = $(shell printf '...%30s: [ \033[31mOFF\033[m ]' $(1))
212 endif
213endef
214
215#
216# Only print out our features if we rebuilt the testcases or if a test failed:
217#
218ifeq ($(test-all-failed), 1)
219 $(foreach feat,$(CORE_FEATURE_TESTS),$(call feature_print,$(feat)))
220 $(info )
221endif
222
223ifeq ($(feature-stackprotector-all), 1) 222ifeq ($(feature-stackprotector-all), 1)
224 CFLAGS += -fstack-protector-all 223 CFLAGS += -fstack-protector-all
225endif 224endif
@@ -264,6 +263,7 @@ ifdef NO_LIBELF
264 NO_DWARF := 1 263 NO_DWARF := 1
265 NO_DEMANGLE := 1 264 NO_DEMANGLE := 1
266 NO_LIBUNWIND := 1 265 NO_LIBUNWIND := 1
266 NO_LIBDW_DWARF_UNWIND := 1
267else 267else
268 ifeq ($(feature-libelf), 0) 268 ifeq ($(feature-libelf), 0)
269 ifeq ($(feature-glibc), 1) 269 ifeq ($(feature-glibc), 1)
@@ -282,13 +282,12 @@ else
282 msg := $(error No gnu/libc-version.h found, please install glibc-dev[el]/glibc-static); 282 msg := $(error No gnu/libc-version.h found, please install glibc-dev[el]/glibc-static);
283 endif 283 endif
284 else 284 else
285 # for linking with debug library, run like: 285 ifndef NO_LIBDW_DWARF_UNWIND
286 # make DEBUG=1 LIBDW_DIR=/opt/libdw/ 286 ifneq ($(feature-libdw-dwarf-unwind),1)
287 ifdef LIBDW_DIR 287 NO_LIBDW_DWARF_UNWIND := 1
288 LIBDW_CFLAGS := -I$(LIBDW_DIR)/include 288 msg := $(warning No libdw DWARF unwind found, Please install elfutils-devel/libdw-dev >= 0.158 and/or set LIBDW_DIR);
289 LIBDW_LDFLAGS := -L$(LIBDW_DIR)/lib 289 endif
290 endif 290 endif
291
292 ifneq ($(feature-dwarf), 1) 291 ifneq ($(feature-dwarf), 1)
293 msg := $(warning No libdw.h found or old libdw.h found or elfutils is older than 0.138, disables dwarf support. Please install new elfutils-devel/libdw-dev); 292 msg := $(warning No libdw.h found or old libdw.h found or elfutils is older than 0.138, disables dwarf support. Please install new elfutils-devel/libdw-dev);
294 NO_DWARF := 1 293 NO_DWARF := 1
@@ -324,25 +323,51 @@ endif # NO_LIBELF
324 323
325ifndef NO_LIBUNWIND 324ifndef NO_LIBUNWIND
326 ifneq ($(feature-libunwind), 1) 325 ifneq ($(feature-libunwind), 1)
327 msg := $(warning No libunwind found, disabling post unwind support. Please install libunwind-dev[el] >= 1.1); 326 msg := $(warning No libunwind found. Please install libunwind-dev[el] >= 1.1 and/or set LIBUNWIND_DIR);
328 NO_LIBUNWIND := 1 327 NO_LIBUNWIND := 1
328 endif
329endif
330
331dwarf-post-unwind := 1
332dwarf-post-unwind-text := BUG
333
334# setup DWARF post unwinder
335ifdef NO_LIBUNWIND
336 ifdef NO_LIBDW_DWARF_UNWIND
337 msg := $(warning Disabling post unwind, no support found.);
338 dwarf-post-unwind := 0
329 else 339 else
330 ifeq ($(ARCH),arm) 340 dwarf-post-unwind-text := libdw
331 $(call feature_check,libunwind-debug-frame) 341 endif
332 ifneq ($(feature-libunwind-debug-frame), 1) 342else
333 msg := $(warning No debug_frame support found in libunwind); 343 dwarf-post-unwind-text := libunwind
334 CFLAGS += -DNO_LIBUNWIND_DEBUG_FRAME 344 # Enable libunwind support by default.
335 endif 345 ifndef NO_LIBDW_DWARF_UNWIND
336 else 346 NO_LIBDW_DWARF_UNWIND := 1
337 # non-ARM has no dwarf_find_debug_frame() function: 347 endif
348endif
349
350ifeq ($(dwarf-post-unwind),1)
351 CFLAGS += -DHAVE_DWARF_UNWIND_SUPPORT
352else
353 NO_DWARF_UNWIND := 1
354endif
355
356ifndef NO_LIBUNWIND
357 ifeq ($(ARCH),arm)
358 $(call feature_check,libunwind-debug-frame)
359 ifneq ($(feature-libunwind-debug-frame), 1)
360 msg := $(warning No debug_frame support found in libunwind);
338 CFLAGS += -DNO_LIBUNWIND_DEBUG_FRAME 361 CFLAGS += -DNO_LIBUNWIND_DEBUG_FRAME
339 endif 362 endif
340 363 else
341 CFLAGS += -DHAVE_LIBUNWIND_SUPPORT 364 # non-ARM has no dwarf_find_debug_frame() function:
342 EXTLIBS += $(LIBUNWIND_LIBS) 365 CFLAGS += -DNO_LIBUNWIND_DEBUG_FRAME
343 CFLAGS += $(LIBUNWIND_CFLAGS) 366 endif
344 LDFLAGS += $(LIBUNWIND_LDFLAGS) 367 CFLAGS += -DHAVE_LIBUNWIND_SUPPORT
345 endif # ifneq ($(feature-libunwind), 1) 368 EXTLIBS += $(LIBUNWIND_LIBS)
369 CFLAGS += $(LIBUNWIND_CFLAGS)
370 LDFLAGS += $(LIBUNWIND_LDFLAGS)
346endif 371endif
347 372
348ifndef NO_LIBAUDIT 373ifndef NO_LIBAUDIT
@@ -602,3 +627,84 @@ ifdef DESTDIR
602plugindir=$(libdir)/traceevent/plugins 627plugindir=$(libdir)/traceevent/plugins
603plugindir_SQ= $(subst ','\'',$(plugindir)) 628plugindir_SQ= $(subst ','\'',$(plugindir))
604endif 629endif
630
631#
632# Print the result of the feature test:
633#
634feature_print_status = $(eval $(feature_print_status_code)) $(info $(MSG))
635
636define feature_print_status_code
637 ifeq ($(feature-$(1)), 1)
638 MSG = $(shell printf '...%30s: [ \033[32mon\033[m ]' $(1))
639 else
640 MSG = $(shell printf '...%30s: [ \033[31mOFF\033[m ]' $(1))
641 endif
642endef
643
644feature_print_var = $(eval $(feature_print_var_code)) $(info $(MSG))
645define feature_print_var_code
646 MSG = $(shell printf '...%30s: %s' $(1) $($(1)))
647endef
648
649feature_print_text = $(eval $(feature_print_text_code)) $(info $(MSG))
650define feature_print_text_code
651 MSG = $(shell printf '...%30s: %s' $(1) $(2))
652endef
653
654PERF_FEATURES := $(foreach feat,$(LIB_FEATURE_TESTS),feature-$(feat)($(feature-$(feat))))
655PERF_FEATURES_FILE := $(shell touch $(OUTPUT)PERF-FEATURES; cat $(OUTPUT)PERF-FEATURES)
656
657ifeq ($(dwarf-post-unwind),1)
658 PERF_FEATURES += dwarf-post-unwind($(dwarf-post-unwind-text))
659endif
660
661# The $(display_lib) controls the default detection message
662# output. It's set if:
663# - detected features differes from stored features from
664# last build (in PERF-FEATURES file)
665# - one of the $(LIB_FEATURE_TESTS) is not detected
666# - VF is enabled
667
668ifneq ("$(PERF_FEATURES)","$(PERF_FEATURES_FILE)")
669 $(shell echo "$(PERF_FEATURES)" > $(OUTPUT)PERF-FEATURES)
670 display_lib := 1
671endif
672
673feature_check = $(eval $(feature_check_code))
674define feature_check_code
675 ifneq ($(feature-$(1)), 1)
676 display_lib := 1
677 endif
678endef
679
680$(foreach feat,$(LIB_FEATURE_TESTS),$(call feature_check,$(feat)))
681
682ifeq ($(VF),1)
683 display_lib := 1
684 display_vf := 1
685endif
686
687ifeq ($(display_lib),1)
688 $(info )
689 $(info Auto-detecting system features:)
690 $(foreach feat,$(LIB_FEATURE_TESTS),$(call feature_print_status,$(feat),))
691
692 ifeq ($(dwarf-post-unwind),1)
693 $(call feature_print_text,"DWARF post unwind library", $(dwarf-post-unwind-text))
694 endif
695endif
696
697ifeq ($(display_vf),1)
698 $(foreach feat,$(VF_FEATURE_TESTS),$(call feature_print_status,$(feat),))
699 $(info )
700 $(call feature_print_var,prefix)
701 $(call feature_print_var,bindir)
702 $(call feature_print_var,libdir)
703 $(call feature_print_var,sysconfdir)
704 $(call feature_print_var,LIBUNWIND_DIR)
705 $(call feature_print_var,LIBDW_DIR)
706endif
707
708ifeq ($(display_lib),1)
709 $(info )
710endif
diff --git a/tools/perf/config/feature-checks/Makefile b/tools/perf/config/feature-checks/Makefile
index 523b7bc10553..2da103c53f89 100644
--- a/tools/perf/config/feature-checks/Makefile
+++ b/tools/perf/config/feature-checks/Makefile
@@ -26,7 +26,8 @@ FILES= \
26 test-libunwind-debug-frame.bin \ 26 test-libunwind-debug-frame.bin \
27 test-on-exit.bin \ 27 test-on-exit.bin \
28 test-stackprotector-all.bin \ 28 test-stackprotector-all.bin \
29 test-timerfd.bin 29 test-timerfd.bin \
30 test-libdw-dwarf-unwind.bin
30 31
31CC := $(CROSS_COMPILE)gcc -MD 32CC := $(CROSS_COMPILE)gcc -MD
32PKG_CONFIG := $(CROSS_COMPILE)pkg-config 33PKG_CONFIG := $(CROSS_COMPILE)pkg-config
@@ -141,6 +142,9 @@ test-backtrace.bin:
141test-timerfd.bin: 142test-timerfd.bin:
142 $(BUILD) 143 $(BUILD)
143 144
145test-libdw-dwarf-unwind.bin:
146 $(BUILD)
147
144-include *.d 148-include *.d
145 149
146############################### 150###############################
diff --git a/tools/perf/config/feature-checks/test-all.c b/tools/perf/config/feature-checks/test-all.c
index 9b8a544155bb..fc37eb3ca17b 100644
--- a/tools/perf/config/feature-checks/test-all.c
+++ b/tools/perf/config/feature-checks/test-all.c
@@ -89,6 +89,10 @@
89# include "test-stackprotector-all.c" 89# include "test-stackprotector-all.c"
90#undef main 90#undef main
91 91
92#define main main_test_libdw_dwarf_unwind
93# include "test-libdw-dwarf-unwind.c"
94#undef main
95
92int main(int argc, char *argv[]) 96int main(int argc, char *argv[])
93{ 97{
94 main_test_libpython(); 98 main_test_libpython();
@@ -111,6 +115,7 @@ int main(int argc, char *argv[])
111 main_test_libnuma(); 115 main_test_libnuma();
112 main_test_timerfd(); 116 main_test_timerfd();
113 main_test_stackprotector_all(); 117 main_test_stackprotector_all();
118 main_test_libdw_dwarf_unwind();
114 119
115 return 0; 120 return 0;
116} 121}
diff --git a/tools/perf/config/feature-checks/test-libdw-dwarf-unwind.c b/tools/perf/config/feature-checks/test-libdw-dwarf-unwind.c
new file mode 100644
index 000000000000..f676a3ff442a
--- /dev/null
+++ b/tools/perf/config/feature-checks/test-libdw-dwarf-unwind.c
@@ -0,0 +1,13 @@
1
2#include <elfutils/libdwfl.h>
3
4int main(void)
5{
6 /*
7 * This function is guarded via: __nonnull_attribute__ (1, 2).
8 * Passing '1' as arguments value. This code is never executed,
9 * only compiled.
10 */
11 dwfl_thread_getframes((void *) 1, (void *) 1, NULL);
12 return 0;
13}
diff --git a/tools/perf/design.txt b/tools/perf/design.txt
index 63a0e6f04a01..a28dca2582aa 100644
--- a/tools/perf/design.txt
+++ b/tools/perf/design.txt
@@ -18,7 +18,7 @@ underlying hardware counters.
18Performance counters are accessed via special file descriptors. 18Performance counters are accessed via special file descriptors.
19There's one file descriptor per virtual counter used. 19There's one file descriptor per virtual counter used.
20 20
21The special file descriptor is opened via the perf_event_open() 21The special file descriptor is opened via the sys_perf_event_open()
22system call: 22system call:
23 23
24 int sys_perf_event_open(struct perf_event_attr *hw_event_uptr, 24 int sys_perf_event_open(struct perf_event_attr *hw_event_uptr,
@@ -82,7 +82,7 @@ machine-specific.
82If 'raw_type' is 0, then the 'type' field says what kind of counter 82If 'raw_type' is 0, then the 'type' field says what kind of counter
83this is, with the following encoding: 83this is, with the following encoding:
84 84
85enum perf_event_types { 85enum perf_type_id {
86 PERF_TYPE_HARDWARE = 0, 86 PERF_TYPE_HARDWARE = 0,
87 PERF_TYPE_SOFTWARE = 1, 87 PERF_TYPE_SOFTWARE = 1,
88 PERF_TYPE_TRACEPOINT = 2, 88 PERF_TYPE_TRACEPOINT = 2,
@@ -95,7 +95,7 @@ specified by 'event_id':
95 * Generalized performance counter event types, used by the hw_event.event_id 95 * Generalized performance counter event types, used by the hw_event.event_id
96 * parameter of the sys_perf_event_open() syscall: 96 * parameter of the sys_perf_event_open() syscall:
97 */ 97 */
98enum hw_event_ids { 98enum perf_hw_id {
99 /* 99 /*
100 * Common hardware events, generalized by the kernel: 100 * Common hardware events, generalized by the kernel:
101 */ 101 */
@@ -129,7 +129,7 @@ software events, selected by 'event_id':
129 * physical and sw events of the kernel (and allow the profiling of them as 129 * physical and sw events of the kernel (and allow the profiling of them as
130 * well): 130 * well):
131 */ 131 */
132enum sw_event_ids { 132enum perf_sw_ids {
133 PERF_COUNT_SW_CPU_CLOCK = 0, 133 PERF_COUNT_SW_CPU_CLOCK = 0,
134 PERF_COUNT_SW_TASK_CLOCK = 1, 134 PERF_COUNT_SW_TASK_CLOCK = 1,
135 PERF_COUNT_SW_PAGE_FAULTS = 2, 135 PERF_COUNT_SW_PAGE_FAULTS = 2,
@@ -230,7 +230,7 @@ these events are recorded in the ring-buffer (see below).
230The 'comm' bit allows tracking of process comm data on process creation. 230The 'comm' bit allows tracking of process comm data on process creation.
231This too is recorded in the ring-buffer (see below). 231This too is recorded in the ring-buffer (see below).
232 232
233The 'pid' parameter to the perf_event_open() system call allows the 233The 'pid' parameter to the sys_perf_event_open() system call allows the
234counter to be specific to a task: 234counter to be specific to a task:
235 235
236 pid == 0: if the pid parameter is zero, the counter is attached to the 236 pid == 0: if the pid parameter is zero, the counter is attached to the
@@ -260,7 +260,7 @@ The 'flags' parameter is currently unused and must be zero.
260 260
261The 'group_fd' parameter allows counter "groups" to be set up. A 261The 'group_fd' parameter allows counter "groups" to be set up. A
262counter group has one counter which is the group "leader". The leader 262counter group has one counter which is the group "leader". The leader
263is created first, with group_fd = -1 in the perf_event_open call 263is created first, with group_fd = -1 in the sys_perf_event_open call
264that creates it. The rest of the group members are created 264that creates it. The rest of the group members are created
265subsequently, with group_fd giving the fd of the group leader. 265subsequently, with group_fd giving the fd of the group leader.
266(A single counter on its own is created with group_fd = -1 and is 266(A single counter on its own is created with group_fd = -1 and is
diff --git a/tools/perf/perf-completion.sh b/tools/perf/perf-completion.sh
index 496e2abb5482..ae3a57694b6b 100644
--- a/tools/perf/perf-completion.sh
+++ b/tools/perf/perf-completion.sh
@@ -123,7 +123,7 @@ __perf_main ()
123 __perfcomp_colon "$evts" "$cur" 123 __perfcomp_colon "$evts" "$cur"
124 # List subcommands for 'perf kvm' 124 # List subcommands for 'perf kvm'
125 elif [[ $prev == "kvm" ]]; then 125 elif [[ $prev == "kvm" ]]; then
126 subcmds="top record report diff buildid-list stat" 126 subcmds=$($cmd $prev --list-cmds)
127 __perfcomp_colon "$subcmds" "$cur" 127 __perfcomp_colon "$subcmds" "$cur"
128 # List long option names 128 # List long option names
129 elif [[ $cur == --* ]]; then 129 elif [[ $cur == --* ]]; then
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index e84fa26bc1be..e18a8b5e6953 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -12,6 +12,9 @@
12#ifndef __NR_perf_event_open 12#ifndef __NR_perf_event_open
13# define __NR_perf_event_open 336 13# define __NR_perf_event_open 336
14#endif 14#endif
15#ifndef __NR_futex
16# define __NR_futex 240
17#endif
15#endif 18#endif
16 19
17#if defined(__x86_64__) 20#if defined(__x86_64__)
@@ -23,6 +26,9 @@
23#ifndef __NR_perf_event_open 26#ifndef __NR_perf_event_open
24# define __NR_perf_event_open 298 27# define __NR_perf_event_open 298
25#endif 28#endif
29#ifndef __NR_futex
30# define __NR_futex 202
31#endif
26#endif 32#endif
27 33
28#ifdef __powerpc__ 34#ifdef __powerpc__
@@ -251,12 +257,14 @@ void pthread__unblock_sigwinch(void);
251enum perf_call_graph_mode { 257enum perf_call_graph_mode {
252 CALLCHAIN_NONE, 258 CALLCHAIN_NONE,
253 CALLCHAIN_FP, 259 CALLCHAIN_FP,
254 CALLCHAIN_DWARF 260 CALLCHAIN_DWARF,
261 CALLCHAIN_MAX
255}; 262};
256 263
257struct record_opts { 264struct record_opts {
258 struct target target; 265 struct target target;
259 int call_graph; 266 int call_graph;
267 bool call_graph_enabled;
260 bool group; 268 bool group;
261 bool inherit_stat; 269 bool inherit_stat;
262 bool no_buffering; 270 bool no_buffering;
diff --git a/tools/perf/tests/builtin-test.c b/tools/perf/tests/builtin-test.c
index 1e67437fb4ca..b11bf8a08430 100644
--- a/tools/perf/tests/builtin-test.c
+++ b/tools/perf/tests/builtin-test.c
@@ -115,6 +115,14 @@ static struct test {
115 .desc = "Test parsing with no sample_id_all bit set", 115 .desc = "Test parsing with no sample_id_all bit set",
116 .func = test__parse_no_sample_id_all, 116 .func = test__parse_no_sample_id_all,
117 }, 117 },
118#if defined(__x86_64__) || defined(__i386__)
119#ifdef HAVE_DWARF_UNWIND_SUPPORT
120 {
121 .desc = "Test dwarf unwind",
122 .func = test__dwarf_unwind,
123 },
124#endif
125#endif
118 { 126 {
119 .func = NULL, 127 .func = NULL,
120 }, 128 },
diff --git a/tools/perf/tests/dwarf-unwind.c b/tools/perf/tests/dwarf-unwind.c
new file mode 100644
index 000000000000..c059ee81c038
--- /dev/null
+++ b/tools/perf/tests/dwarf-unwind.c
@@ -0,0 +1,144 @@
1#include <linux/compiler.h>
2#include <sys/types.h>
3#include <unistd.h>
4#include "tests.h"
5#include "debug.h"
6#include "machine.h"
7#include "event.h"
8#include "unwind.h"
9#include "perf_regs.h"
10#include "map.h"
11#include "thread.h"
12
13static int mmap_handler(struct perf_tool *tool __maybe_unused,
14 union perf_event *event,
15 struct perf_sample *sample __maybe_unused,
16 struct machine *machine)
17{
18 return machine__process_mmap_event(machine, event, NULL);
19}
20
21static int init_live_machine(struct machine *machine)
22{
23 union perf_event event;
24 pid_t pid = getpid();
25
26 return perf_event__synthesize_mmap_events(NULL, &event, pid, pid,
27 mmap_handler, machine, true);
28}
29
30#define MAX_STACK 6
31
32static int unwind_entry(struct unwind_entry *entry, void *arg)
33{
34 unsigned long *cnt = (unsigned long *) arg;
35 char *symbol = entry->sym ? entry->sym->name : NULL;
36 static const char *funcs[MAX_STACK] = {
37 "test__arch_unwind_sample",
38 "unwind_thread",
39 "krava_3",
40 "krava_2",
41 "krava_1",
42 "test__dwarf_unwind"
43 };
44
45 if (*cnt >= MAX_STACK) {
46 pr_debug("failed: crossed the max stack value %d\n", MAX_STACK);
47 return -1;
48 }
49
50 if (!symbol) {
51 pr_debug("failed: got unresolved address 0x%" PRIx64 "\n",
52 entry->ip);
53 return -1;
54 }
55
56 pr_debug("got: %s 0x%" PRIx64 "\n", symbol, entry->ip);
57 return strcmp((const char *) symbol, funcs[(*cnt)++]);
58}
59
60__attribute__ ((noinline))
61static int unwind_thread(struct thread *thread, struct machine *machine)
62{
63 struct perf_sample sample;
64 unsigned long cnt = 0;
65 int err = -1;
66
67 memset(&sample, 0, sizeof(sample));
68
69 if (test__arch_unwind_sample(&sample, thread)) {
70 pr_debug("failed to get unwind sample\n");
71 goto out;
72 }
73
74 err = unwind__get_entries(unwind_entry, &cnt, machine, thread,
75 &sample, MAX_STACK);
76 if (err)
77 pr_debug("unwind failed\n");
78 else if (cnt != MAX_STACK) {
79 pr_debug("got wrong number of stack entries %lu != %d\n",
80 cnt, MAX_STACK);
81 err = -1;
82 }
83
84 out:
85 free(sample.user_stack.data);
86 free(sample.user_regs.regs);
87 return err;
88}
89
90__attribute__ ((noinline))
91static int krava_3(struct thread *thread, struct machine *machine)
92{
93 return unwind_thread(thread, machine);
94}
95
96__attribute__ ((noinline))
97static int krava_2(struct thread *thread, struct machine *machine)
98{
99 return krava_3(thread, machine);
100}
101
102__attribute__ ((noinline))
103static int krava_1(struct thread *thread, struct machine *machine)
104{
105 return krava_2(thread, machine);
106}
107
108int test__dwarf_unwind(void)
109{
110 struct machines machines;
111 struct machine *machine;
112 struct thread *thread;
113 int err = -1;
114
115 machines__init(&machines);
116
117 machine = machines__find(&machines, HOST_KERNEL_ID);
118 if (!machine) {
119 pr_err("Could not get machine\n");
120 return -1;
121 }
122
123 if (init_live_machine(machine)) {
124 pr_err("Could not init machine\n");
125 goto out;
126 }
127
128 if (verbose > 1)
129 machine__fprintf(machine, stderr);
130
131 thread = machine__find_thread(machine, getpid(), getpid());
132 if (!thread) {
133 pr_err("Could not get thread\n");
134 goto out;
135 }
136
137 err = krava_1(thread, machine);
138
139 out:
140 machine__delete_threads(machine);
141 machine__exit(machine);
142 machines__exit(&machines);
143 return err;
144}
diff --git a/tools/perf/tests/hists_link.c b/tools/perf/tests/hists_link.c
index 2b6519e0e36f..7ccbc7b6ae77 100644
--- a/tools/perf/tests/hists_link.c
+++ b/tools/perf/tests/hists_link.c
@@ -101,6 +101,7 @@ static struct machine *setup_fake_machine(struct machines *machines)
101 .mmap = { 101 .mmap = {
102 .header = { .misc = PERF_RECORD_MISC_USER, }, 102 .header = { .misc = PERF_RECORD_MISC_USER, },
103 .pid = fake_mmap_info[i].pid, 103 .pid = fake_mmap_info[i].pid,
104 .tid = fake_mmap_info[i].pid,
104 .start = fake_mmap_info[i].start, 105 .start = fake_mmap_info[i].start,
105 .len = 0x1000ULL, 106 .len = 0x1000ULL,
106 .pgoff = 0ULL, 107 .pgoff = 0ULL,
diff --git a/tools/perf/tests/make b/tools/perf/tests/make
index 00544b8b644b..5daeae1cb4c0 100644
--- a/tools/perf/tests/make
+++ b/tools/perf/tests/make
@@ -27,6 +27,7 @@ make_no_ui := NO_NEWT=1 NO_SLANG=1 NO_GTK2=1
27make_no_demangle := NO_DEMANGLE=1 27make_no_demangle := NO_DEMANGLE=1
28make_no_libelf := NO_LIBELF=1 28make_no_libelf := NO_LIBELF=1
29make_no_libunwind := NO_LIBUNWIND=1 29make_no_libunwind := NO_LIBUNWIND=1
30make_no_libdw_dwarf_unwind := NO_LIBDW_DWARF_UNWIND=1
30make_no_backtrace := NO_BACKTRACE=1 31make_no_backtrace := NO_BACKTRACE=1
31make_no_libnuma := NO_LIBNUMA=1 32make_no_libnuma := NO_LIBNUMA=1
32make_no_libaudit := NO_LIBAUDIT=1 33make_no_libaudit := NO_LIBAUDIT=1
@@ -35,8 +36,9 @@ make_tags := tags
35make_cscope := cscope 36make_cscope := cscope
36make_help := help 37make_help := help
37make_doc := doc 38make_doc := doc
38make_perf_o := perf.o 39make_perf_o := perf.o
39make_util_map_o := util/map.o 40make_util_map_o := util/map.o
41make_util_pmu_bison_o := util/pmu-bison.o
40make_install := install 42make_install := install
41make_install_bin := install-bin 43make_install_bin := install-bin
42make_install_doc := install-doc 44make_install_doc := install-doc
@@ -49,6 +51,7 @@ make_install_pdf := install-pdf
49make_minimal := NO_LIBPERL=1 NO_LIBPYTHON=1 NO_NEWT=1 NO_GTK2=1 51make_minimal := NO_LIBPERL=1 NO_LIBPYTHON=1 NO_NEWT=1 NO_GTK2=1
50make_minimal += NO_DEMANGLE=1 NO_LIBELF=1 NO_LIBUNWIND=1 NO_BACKTRACE=1 52make_minimal += NO_DEMANGLE=1 NO_LIBELF=1 NO_LIBUNWIND=1 NO_BACKTRACE=1
51make_minimal += NO_LIBNUMA=1 NO_LIBAUDIT=1 NO_LIBBIONIC=1 53make_minimal += NO_LIBNUMA=1 NO_LIBAUDIT=1 NO_LIBBIONIC=1
54make_minimal += NO_LIBDW_DWARF_UNWIND=1
52 55
53# $(run) contains all available tests 56# $(run) contains all available tests
54run := make_pure 57run := make_pure
@@ -65,6 +68,7 @@ run += make_no_ui
65run += make_no_demangle 68run += make_no_demangle
66run += make_no_libelf 69run += make_no_libelf
67run += make_no_libunwind 70run += make_no_libunwind
71run += make_no_libdw_dwarf_unwind
68run += make_no_backtrace 72run += make_no_backtrace
69run += make_no_libnuma 73run += make_no_libnuma
70run += make_no_libaudit 74run += make_no_libaudit
@@ -73,6 +77,7 @@ run += make_help
73run += make_doc 77run += make_doc
74run += make_perf_o 78run += make_perf_o
75run += make_util_map_o 79run += make_util_map_o
80run += make_util_pmu_bison_o
76run += make_install 81run += make_install
77run += make_install_bin 82run += make_install_bin
78# FIXME 'install-*' commented out till they're fixed 83# FIXME 'install-*' commented out till they're fixed
@@ -113,8 +118,9 @@ test_make_doc_O := $(test_ok)
113 118
114test_make_python_perf_so := test -f $(PERF)/python/perf.so 119test_make_python_perf_so := test -f $(PERF)/python/perf.so
115 120
116test_make_perf_o := test -f $(PERF)/perf.o 121test_make_perf_o := test -f $(PERF)/perf.o
117test_make_util_map_o := test -f $(PERF)/util/map.o 122test_make_util_map_o := test -f $(PERF)/util/map.o
123test_make_util_pmu_bison_o := test -f $(PERF)/util/pmu-bison.o
118 124
119define test_dest_files 125define test_dest_files
120 for file in $(1); do \ 126 for file in $(1); do \
@@ -167,13 +173,10 @@ test_make_install_info_O := $(test_ok)
167test_make_install_pdf := $(test_ok) 173test_make_install_pdf := $(test_ok)
168test_make_install_pdf_O := $(test_ok) 174test_make_install_pdf_O := $(test_ok)
169 175
170# Kbuild tests only 176test_make_python_perf_so_O := test -f $$TMP_O/python/perf.so
171#test_make_python_perf_so_O := test -f $$TMP/tools/perf/python/perf.so 177test_make_perf_o_O := test -f $$TMP_O/perf.o
172#test_make_perf_o_O := test -f $$TMP/tools/perf/perf.o 178test_make_util_map_o_O := test -f $$TMP_O/util/map.o
173#test_make_util_map_o_O := test -f $$TMP/tools/perf/util/map.o 179test_make_util_pmu_bison_o_O := test -f $$TMP_O/util/pmu-bison.o
174
175test_make_perf_o_O := true
176test_make_util_map_o_O := true
177 180
178test_default = test -x $(PERF)/perf 181test_default = test -x $(PERF)/perf
179test = $(if $(test_$1),$(test_$1),$(test_default)) 182test = $(if $(test_$1),$(test_$1),$(test_default))
diff --git a/tools/perf/tests/parse-events.c b/tools/perf/tests/parse-events.c
index 4db0ae617d70..8605ff5572ae 100644
--- a/tools/perf/tests/parse-events.c
+++ b/tools/perf/tests/parse-events.c
@@ -2,7 +2,7 @@
2#include "parse-events.h" 2#include "parse-events.h"
3#include "evsel.h" 3#include "evsel.h"
4#include "evlist.h" 4#include "evlist.h"
5#include "fs.h" 5#include <api/fs/fs.h>
6#include <api/fs/debugfs.h> 6#include <api/fs/debugfs.h>
7#include "tests.h" 7#include "tests.h"
8#include <linux/hw_breakpoint.h> 8#include <linux/hw_breakpoint.h>
diff --git a/tools/perf/tests/sample-parsing.c b/tools/perf/tests/sample-parsing.c
index 1b677202638d..0014d3c8c21c 100644
--- a/tools/perf/tests/sample-parsing.c
+++ b/tools/perf/tests/sample-parsing.c
@@ -22,8 +22,8 @@
22} while (0) 22} while (0)
23 23
24static bool samples_same(const struct perf_sample *s1, 24static bool samples_same(const struct perf_sample *s1,
25 const struct perf_sample *s2, u64 type, u64 regs_user, 25 const struct perf_sample *s2,
26 u64 read_format) 26 u64 type, u64 read_format)
27{ 27{
28 size_t i; 28 size_t i;
29 29
@@ -95,8 +95,9 @@ static bool samples_same(const struct perf_sample *s1,
95 } 95 }
96 96
97 if (type & PERF_SAMPLE_REGS_USER) { 97 if (type & PERF_SAMPLE_REGS_USER) {
98 size_t sz = hweight_long(regs_user) * sizeof(u64); 98 size_t sz = hweight_long(s1->user_regs.mask) * sizeof(u64);
99 99
100 COMP(user_regs.mask);
100 COMP(user_regs.abi); 101 COMP(user_regs.abi);
101 if (s1->user_regs.abi && 102 if (s1->user_regs.abi &&
102 (!s1->user_regs.regs || !s2->user_regs.regs || 103 (!s1->user_regs.regs || !s2->user_regs.regs ||
@@ -174,6 +175,7 @@ static int do_test(u64 sample_type, u64 sample_regs_user, u64 read_format)
174 .branch_stack = &branch_stack.branch_stack, 175 .branch_stack = &branch_stack.branch_stack,
175 .user_regs = { 176 .user_regs = {
176 .abi = PERF_SAMPLE_REGS_ABI_64, 177 .abi = PERF_SAMPLE_REGS_ABI_64,
178 .mask = sample_regs_user,
177 .regs = user_regs, 179 .regs = user_regs,
178 }, 180 },
179 .user_stack = { 181 .user_stack = {
@@ -201,8 +203,7 @@ static int do_test(u64 sample_type, u64 sample_regs_user, u64 read_format)
201 sample.read.one.id = 99; 203 sample.read.one.id = 99;
202 } 204 }
203 205
204 sz = perf_event__sample_event_size(&sample, sample_type, 206 sz = perf_event__sample_event_size(&sample, sample_type, read_format);
205 sample_regs_user, read_format);
206 bufsz = sz + 4096; /* Add a bit for overrun checking */ 207 bufsz = sz + 4096; /* Add a bit for overrun checking */
207 event = malloc(bufsz); 208 event = malloc(bufsz);
208 if (!event) { 209 if (!event) {
@@ -215,8 +216,7 @@ static int do_test(u64 sample_type, u64 sample_regs_user, u64 read_format)
215 event->header.misc = 0; 216 event->header.misc = 0;
216 event->header.size = sz; 217 event->header.size = sz;
217 218
218 err = perf_event__synthesize_sample(event, sample_type, 219 err = perf_event__synthesize_sample(event, sample_type, read_format,
219 sample_regs_user, read_format,
220 &sample, false); 220 &sample, false);
221 if (err) { 221 if (err) {
222 pr_debug("%s failed for sample_type %#"PRIx64", error %d\n", 222 pr_debug("%s failed for sample_type %#"PRIx64", error %d\n",
@@ -244,8 +244,7 @@ static int do_test(u64 sample_type, u64 sample_regs_user, u64 read_format)
244 goto out_free; 244 goto out_free;
245 } 245 }
246 246
247 if (!samples_same(&sample, &sample_out, sample_type, 247 if (!samples_same(&sample, &sample_out, sample_type, read_format)) {
248 sample_regs_user, read_format)) {
249 pr_debug("parsing failed for sample_type %#"PRIx64"\n", 248 pr_debug("parsing failed for sample_type %#"PRIx64"\n",
250 sample_type); 249 sample_type);
251 goto out_free; 250 goto out_free;
diff --git a/tools/perf/tests/tests.h b/tools/perf/tests/tests.h
index e0ac713857ba..a24795ca002d 100644
--- a/tools/perf/tests/tests.h
+++ b/tools/perf/tests/tests.h
@@ -40,5 +40,14 @@ int test__code_reading(void);
40int test__sample_parsing(void); 40int test__sample_parsing(void);
41int test__keep_tracking(void); 41int test__keep_tracking(void);
42int test__parse_no_sample_id_all(void); 42int test__parse_no_sample_id_all(void);
43int test__dwarf_unwind(void);
43 44
45#if defined(__x86_64__) || defined(__i386__)
46#ifdef HAVE_DWARF_UNWIND_SUPPORT
47struct thread;
48struct perf_sample;
49int test__arch_unwind_sample(struct perf_sample *sample,
50 struct thread *thread);
51#endif
52#endif
44#endif /* TESTS_H */ 53#endif /* TESTS_H */
diff --git a/tools/perf/ui/browsers/hists.c b/tools/perf/ui/browsers/hists.c
index b720b92eba6e..7ec871af3f6f 100644
--- a/tools/perf/ui/browsers/hists.c
+++ b/tools/perf/ui/browsers/hists.c
@@ -587,95 +587,52 @@ struct hpp_arg {
587 bool current_entry; 587 bool current_entry;
588}; 588};
589 589
590static int __hpp__color_callchain(struct hpp_arg *arg) 590static int __hpp__overhead_callback(struct perf_hpp *hpp, bool front)
591{ 591{
592 if (!symbol_conf.use_callchain)
593 return 0;
594
595 slsmg_printf("%c ", arg->folded_sign);
596 return 2;
597}
598
599static int __hpp__color_fmt(struct perf_hpp *hpp, struct hist_entry *he,
600 u64 (*get_field)(struct hist_entry *),
601 int (*callchain_cb)(struct hpp_arg *))
602{
603 int ret = 0;
604 double percent = 0.0;
605 struct hists *hists = he->hists;
606 struct hpp_arg *arg = hpp->ptr; 592 struct hpp_arg *arg = hpp->ptr;
607 593
608 if (hists->stats.total_period) 594 if (arg->current_entry && arg->b->navkeypressed)
609 percent = 100.0 * get_field(he) / hists->stats.total_period; 595 ui_browser__set_color(arg->b, HE_COLORSET_SELECTED);
610 596 else
611 ui_browser__set_percent_color(arg->b, percent, arg->current_entry); 597 ui_browser__set_color(arg->b, HE_COLORSET_NORMAL);
612
613 if (callchain_cb)
614 ret += callchain_cb(arg);
615
616 ret += scnprintf(hpp->buf, hpp->size, "%6.2f%%", percent);
617 slsmg_printf("%s", hpp->buf);
618
619 if (symbol_conf.event_group) {
620 int prev_idx, idx_delta;
621 struct perf_evsel *evsel = hists_to_evsel(hists);
622 struct hist_entry *pair;
623 int nr_members = evsel->nr_members;
624
625 if (nr_members <= 1)
626 goto out;
627 598
628 prev_idx = perf_evsel__group_idx(evsel); 599 if (front) {
600 if (!symbol_conf.use_callchain)
601 return 0;
629 602
630 list_for_each_entry(pair, &he->pairs.head, pairs.node) { 603 slsmg_printf("%c ", arg->folded_sign);
631 u64 period = get_field(pair); 604 return 2;
632 u64 total = pair->hists->stats.total_period; 605 }
633 606
634 if (!total) 607 return 0;
635 continue; 608}
636 609
637 evsel = hists_to_evsel(pair->hists); 610static int __hpp__color_callback(struct perf_hpp *hpp, bool front __maybe_unused)
638 idx_delta = perf_evsel__group_idx(evsel) - prev_idx - 1; 611{
612 struct hpp_arg *arg = hpp->ptr;
639 613
640 while (idx_delta--) { 614 if (!arg->current_entry || !arg->b->navkeypressed)
641 /* 615 ui_browser__set_color(arg->b, HE_COLORSET_NORMAL);
642 * zero-fill group members in the middle which 616 return 0;
643 * have no sample 617}
644 */
645 ui_browser__set_percent_color(arg->b, 0.0,
646 arg->current_entry);
647 ret += scnprintf(hpp->buf, hpp->size,
648 " %6.2f%%", 0.0);
649 slsmg_printf("%s", hpp->buf);
650 }
651 618
652 percent = 100.0 * period / total; 619static int __hpp__slsmg_color_printf(struct perf_hpp *hpp, const char *fmt, ...)
653 ui_browser__set_percent_color(arg->b, percent, 620{
654 arg->current_entry); 621 struct hpp_arg *arg = hpp->ptr;
655 ret += scnprintf(hpp->buf, hpp->size, 622 int ret;
656 " %6.2f%%", percent); 623 va_list args;
657 slsmg_printf("%s", hpp->buf); 624 double percent;
658 625
659 prev_idx = perf_evsel__group_idx(evsel); 626 va_start(args, fmt);
660 } 627 percent = va_arg(args, double);
628 va_end(args);
661 629
662 idx_delta = nr_members - prev_idx - 1; 630 ui_browser__set_percent_color(arg->b, percent, arg->current_entry);
663 631
664 while (idx_delta--) { 632 ret = scnprintf(hpp->buf, hpp->size, fmt, percent);
665 /* 633 slsmg_printf("%s", hpp->buf);
666 * zero-fill group members at last which have no sample
667 */
668 ui_browser__set_percent_color(arg->b, 0.0,
669 arg->current_entry);
670 ret += scnprintf(hpp->buf, hpp->size,
671 " %6.2f%%", 0.0);
672 slsmg_printf("%s", hpp->buf);
673 }
674 }
675out:
676 if (!arg->current_entry || !arg->b->navkeypressed)
677 ui_browser__set_color(arg->b, HE_COLORSET_NORMAL);
678 634
635 advance_hpp(hpp, ret);
679 return ret; 636 return ret;
680} 637}
681 638
@@ -690,14 +647,15 @@ hist_browser__hpp_color_##_type(struct perf_hpp_fmt *fmt __maybe_unused,\
690 struct perf_hpp *hpp, \ 647 struct perf_hpp *hpp, \
691 struct hist_entry *he) \ 648 struct hist_entry *he) \
692{ \ 649{ \
693 return __hpp__color_fmt(hpp, he, __hpp_get_##_field, _cb); \ 650 return __hpp__fmt(hpp, he, __hpp_get_##_field, _cb, " %6.2f%%", \
651 __hpp__slsmg_color_printf, true); \
694} 652}
695 653
696__HPP_COLOR_PERCENT_FN(overhead, period, __hpp__color_callchain) 654__HPP_COLOR_PERCENT_FN(overhead, period, __hpp__overhead_callback)
697__HPP_COLOR_PERCENT_FN(overhead_sys, period_sys, NULL) 655__HPP_COLOR_PERCENT_FN(overhead_sys, period_sys, __hpp__color_callback)
698__HPP_COLOR_PERCENT_FN(overhead_us, period_us, NULL) 656__HPP_COLOR_PERCENT_FN(overhead_us, period_us, __hpp__color_callback)
699__HPP_COLOR_PERCENT_FN(overhead_guest_sys, period_guest_sys, NULL) 657__HPP_COLOR_PERCENT_FN(overhead_guest_sys, period_guest_sys, __hpp__color_callback)
700__HPP_COLOR_PERCENT_FN(overhead_guest_us, period_guest_us, NULL) 658__HPP_COLOR_PERCENT_FN(overhead_guest_us, period_guest_us, __hpp__color_callback)
701 659
702#undef __HPP_COLOR_PERCENT_FN 660#undef __HPP_COLOR_PERCENT_FN
703 661
diff --git a/tools/perf/ui/gtk/hists.c b/tools/perf/ui/gtk/hists.c
index 5b95c44f3435..e395ef9b0ae0 100644
--- a/tools/perf/ui/gtk/hists.c
+++ b/tools/perf/ui/gtk/hists.c
@@ -8,16 +8,24 @@
8 8
9#define MAX_COLUMNS 32 9#define MAX_COLUMNS 32
10 10
11static int __percent_color_snprintf(char *buf, size_t size, double percent) 11static int __percent_color_snprintf(struct perf_hpp *hpp, const char *fmt, ...)
12{ 12{
13 int ret = 0; 13 int ret = 0;
14 va_list args;
15 double percent;
14 const char *markup; 16 const char *markup;
17 char *buf = hpp->buf;
18 size_t size = hpp->size;
19
20 va_start(args, fmt);
21 percent = va_arg(args, double);
22 va_end(args);
15 23
16 markup = perf_gtk__get_percent_color(percent); 24 markup = perf_gtk__get_percent_color(percent);
17 if (markup) 25 if (markup)
18 ret += scnprintf(buf, size, markup); 26 ret += scnprintf(buf, size, markup);
19 27
20 ret += scnprintf(buf + ret, size - ret, " %6.2f%%", percent); 28 ret += scnprintf(buf + ret, size - ret, fmt, percent);
21 29
22 if (markup) 30 if (markup)
23 ret += scnprintf(buf + ret, size - ret, "</span>"); 31 ret += scnprintf(buf + ret, size - ret, "</span>");
@@ -25,66 +33,6 @@ static int __percent_color_snprintf(char *buf, size_t size, double percent)
25 return ret; 33 return ret;
26} 34}
27 35
28
29static int __hpp__color_fmt(struct perf_hpp *hpp, struct hist_entry *he,
30 u64 (*get_field)(struct hist_entry *))
31{
32 int ret;
33 double percent = 0.0;
34 struct hists *hists = he->hists;
35 struct perf_evsel *evsel = hists_to_evsel(hists);
36
37 if (hists->stats.total_period)
38 percent = 100.0 * get_field(he) / hists->stats.total_period;
39
40 ret = __percent_color_snprintf(hpp->buf, hpp->size, percent);
41
42 if (perf_evsel__is_group_event(evsel)) {
43 int prev_idx, idx_delta;
44 struct hist_entry *pair;
45 int nr_members = evsel->nr_members;
46
47 prev_idx = perf_evsel__group_idx(evsel);
48
49 list_for_each_entry(pair, &he->pairs.head, pairs.node) {
50 u64 period = get_field(pair);
51 u64 total = pair->hists->stats.total_period;
52
53 evsel = hists_to_evsel(pair->hists);
54 idx_delta = perf_evsel__group_idx(evsel) - prev_idx - 1;
55
56 while (idx_delta--) {
57 /*
58 * zero-fill group members in the middle which
59 * have no sample
60 */
61 ret += __percent_color_snprintf(hpp->buf + ret,
62 hpp->size - ret,
63 0.0);
64 }
65
66 percent = 100.0 * period / total;
67 ret += __percent_color_snprintf(hpp->buf + ret,
68 hpp->size - ret,
69 percent);
70
71 prev_idx = perf_evsel__group_idx(evsel);
72 }
73
74 idx_delta = nr_members - prev_idx - 1;
75
76 while (idx_delta--) {
77 /*
78 * zero-fill group members at last which have no sample
79 */
80 ret += __percent_color_snprintf(hpp->buf + ret,
81 hpp->size - ret,
82 0.0);
83 }
84 }
85 return ret;
86}
87
88#define __HPP_COLOR_PERCENT_FN(_type, _field) \ 36#define __HPP_COLOR_PERCENT_FN(_type, _field) \
89static u64 he_get_##_field(struct hist_entry *he) \ 37static u64 he_get_##_field(struct hist_entry *he) \
90{ \ 38{ \
@@ -95,7 +43,8 @@ static int perf_gtk__hpp_color_##_type(struct perf_hpp_fmt *fmt __maybe_unused,
95 struct perf_hpp *hpp, \ 43 struct perf_hpp *hpp, \
96 struct hist_entry *he) \ 44 struct hist_entry *he) \
97{ \ 45{ \
98 return __hpp__color_fmt(hpp, he, he_get_##_field); \ 46 return __hpp__fmt(hpp, he, he_get_##_field, NULL, " %6.2f%%", \
47 __percent_color_snprintf, true); \
99} 48}
100 49
101__HPP_COLOR_PERCENT_FN(overhead, period) 50__HPP_COLOR_PERCENT_FN(overhead, period)
@@ -216,7 +165,6 @@ static void perf_gtk__show_hists(GtkWidget *window, struct hists *hists,
216 struct perf_hpp hpp = { 165 struct perf_hpp hpp = {
217 .buf = s, 166 .buf = s,
218 .size = sizeof(s), 167 .size = sizeof(s),
219 .ptr = hists_to_evsel(hists),
220 }; 168 };
221 169
222 nr_cols = 0; 170 nr_cols = 0;
@@ -243,7 +191,7 @@ static void perf_gtk__show_hists(GtkWidget *window, struct hists *hists,
243 col_idx = 0; 191 col_idx = 0;
244 192
245 perf_hpp__for_each_format(fmt) { 193 perf_hpp__for_each_format(fmt) {
246 fmt->header(fmt, &hpp); 194 fmt->header(fmt, &hpp, hists_to_evsel(hists));
247 195
248 gtk_tree_view_insert_column_with_attributes(GTK_TREE_VIEW(view), 196 gtk_tree_view_insert_column_with_attributes(GTK_TREE_VIEW(view),
249 -1, ltrim(s), 197 -1, ltrim(s),
diff --git a/tools/perf/ui/hist.c b/tools/perf/ui/hist.c
index 78f4c92e9b73..0f403b83e9d1 100644
--- a/tools/perf/ui/hist.c
+++ b/tools/perf/ui/hist.c
@@ -8,16 +8,27 @@
8 8
9/* hist period print (hpp) functions */ 9/* hist period print (hpp) functions */
10 10
11typedef int (*hpp_snprint_fn)(char *buf, size_t size, const char *fmt, ...); 11#define hpp__call_print_fn(hpp, fn, fmt, ...) \
12 12({ \
13static int __hpp__fmt(struct perf_hpp *hpp, struct hist_entry *he, 13 int __ret = fn(hpp, fmt, ##__VA_ARGS__); \
14 u64 (*get_field)(struct hist_entry *), 14 advance_hpp(hpp, __ret); \
15 const char *fmt, hpp_snprint_fn print_fn, 15 __ret; \
16 bool fmt_percent) 16})
17
18int __hpp__fmt(struct perf_hpp *hpp, struct hist_entry *he,
19 hpp_field_fn get_field, hpp_callback_fn callback,
20 const char *fmt, hpp_snprint_fn print_fn, bool fmt_percent)
17{ 21{
18 int ret; 22 int ret = 0;
19 struct hists *hists = he->hists; 23 struct hists *hists = he->hists;
20 struct perf_evsel *evsel = hists_to_evsel(hists); 24 struct perf_evsel *evsel = hists_to_evsel(hists);
25 char *buf = hpp->buf;
26 size_t size = hpp->size;
27
28 if (callback) {
29 ret = callback(hpp, true);
30 advance_hpp(hpp, ret);
31 }
21 32
22 if (fmt_percent) { 33 if (fmt_percent) {
23 double percent = 0.0; 34 double percent = 0.0;
@@ -26,9 +37,9 @@ static int __hpp__fmt(struct perf_hpp *hpp, struct hist_entry *he,
26 percent = 100.0 * get_field(he) / 37 percent = 100.0 * get_field(he) /
27 hists->stats.total_period; 38 hists->stats.total_period;
28 39
29 ret = print_fn(hpp->buf, hpp->size, fmt, percent); 40 ret += hpp__call_print_fn(hpp, print_fn, fmt, percent);
30 } else 41 } else
31 ret = print_fn(hpp->buf, hpp->size, fmt, get_field(he)); 42 ret += hpp__call_print_fn(hpp, print_fn, fmt, get_field(he));
32 43
33 if (perf_evsel__is_group_event(evsel)) { 44 if (perf_evsel__is_group_event(evsel)) {
34 int prev_idx, idx_delta; 45 int prev_idx, idx_delta;
@@ -52,16 +63,22 @@ static int __hpp__fmt(struct perf_hpp *hpp, struct hist_entry *he,
52 * zero-fill group members in the middle which 63 * zero-fill group members in the middle which
53 * have no sample 64 * have no sample
54 */ 65 */
55 ret += print_fn(hpp->buf + ret, hpp->size - ret, 66 if (fmt_percent) {
56 fmt, 0); 67 ret += hpp__call_print_fn(hpp, print_fn,
68 fmt, 0.0);
69 } else {
70 ret += hpp__call_print_fn(hpp, print_fn,
71 fmt, 0ULL);
72 }
57 } 73 }
58 74
59 if (fmt_percent) 75 if (fmt_percent) {
60 ret += print_fn(hpp->buf + ret, hpp->size - ret, 76 ret += hpp__call_print_fn(hpp, print_fn, fmt,
61 fmt, 100.0 * period / total); 77 100.0 * period / total);
62 else 78 } else {
63 ret += print_fn(hpp->buf + ret, hpp->size - ret, 79 ret += hpp__call_print_fn(hpp, print_fn, fmt,
64 fmt, period); 80 period);
81 }
65 82
66 prev_idx = perf_evsel__group_idx(evsel); 83 prev_idx = perf_evsel__group_idx(evsel);
67 } 84 }
@@ -72,41 +89,87 @@ static int __hpp__fmt(struct perf_hpp *hpp, struct hist_entry *he,
72 /* 89 /*
73 * zero-fill group members at last which have no sample 90 * zero-fill group members at last which have no sample
74 */ 91 */
75 ret += print_fn(hpp->buf + ret, hpp->size - ret, 92 if (fmt_percent) {
76 fmt, 0); 93 ret += hpp__call_print_fn(hpp, print_fn,
94 fmt, 0.0);
95 } else {
96 ret += hpp__call_print_fn(hpp, print_fn,
97 fmt, 0ULL);
98 }
77 } 99 }
78 } 100 }
101
102 if (callback) {
103 int __ret = callback(hpp, false);
104
105 advance_hpp(hpp, __ret);
106 ret += __ret;
107 }
108
109 /*
110 * Restore original buf and size as it's where caller expects
111 * the result will be saved.
112 */
113 hpp->buf = buf;
114 hpp->size = size;
115
79 return ret; 116 return ret;
80} 117}
81 118
82#define __HPP_HEADER_FN(_type, _str, _min_width, _unit_width) \ 119#define __HPP_HEADER_FN(_type, _str, _min_width, _unit_width) \
83static int hpp__header_##_type(struct perf_hpp_fmt *fmt __maybe_unused, \ 120static int hpp__header_##_type(struct perf_hpp_fmt *fmt __maybe_unused, \
84 struct perf_hpp *hpp) \ 121 struct perf_hpp *hpp, \
122 struct perf_evsel *evsel) \
85{ \ 123{ \
86 int len = _min_width; \ 124 int len = _min_width; \
87 \ 125 \
88 if (symbol_conf.event_group) { \ 126 if (symbol_conf.event_group) \
89 struct perf_evsel *evsel = hpp->ptr; \
90 \
91 len = max(len, evsel->nr_members * _unit_width); \ 127 len = max(len, evsel->nr_members * _unit_width); \
92 } \ 128 \
93 return scnprintf(hpp->buf, hpp->size, "%*s", len, _str); \ 129 return scnprintf(hpp->buf, hpp->size, "%*s", len, _str); \
94} 130}
95 131
96#define __HPP_WIDTH_FN(_type, _min_width, _unit_width) \ 132#define __HPP_WIDTH_FN(_type, _min_width, _unit_width) \
97static int hpp__width_##_type(struct perf_hpp_fmt *fmt __maybe_unused, \ 133static int hpp__width_##_type(struct perf_hpp_fmt *fmt __maybe_unused, \
98 struct perf_hpp *hpp __maybe_unused) \ 134 struct perf_hpp *hpp __maybe_unused, \
135 struct perf_evsel *evsel) \
99{ \ 136{ \
100 int len = _min_width; \ 137 int len = _min_width; \
101 \ 138 \
102 if (symbol_conf.event_group) { \ 139 if (symbol_conf.event_group) \
103 struct perf_evsel *evsel = hpp->ptr; \
104 \
105 len = max(len, evsel->nr_members * _unit_width); \ 140 len = max(len, evsel->nr_members * _unit_width); \
106 } \ 141 \
107 return len; \ 142 return len; \
108} 143}
109 144
145static int hpp_color_scnprintf(struct perf_hpp *hpp, const char *fmt, ...)
146{
147 va_list args;
148 ssize_t ssize = hpp->size;
149 double percent;
150 int ret;
151
152 va_start(args, fmt);
153 percent = va_arg(args, double);
154 ret = value_color_snprintf(hpp->buf, hpp->size, fmt, percent);
155 va_end(args);
156
157 return (ret >= ssize) ? (ssize - 1) : ret;
158}
159
160static int hpp_entry_scnprintf(struct perf_hpp *hpp, const char *fmt, ...)
161{
162 va_list args;
163 ssize_t ssize = hpp->size;
164 int ret;
165
166 va_start(args, fmt);
167 ret = vsnprintf(hpp->buf, hpp->size, fmt, args);
168 va_end(args);
169
170 return (ret >= ssize) ? (ssize - 1) : ret;
171}
172
110#define __HPP_COLOR_PERCENT_FN(_type, _field) \ 173#define __HPP_COLOR_PERCENT_FN(_type, _field) \
111static u64 he_get_##_field(struct hist_entry *he) \ 174static u64 he_get_##_field(struct hist_entry *he) \
112{ \ 175{ \
@@ -116,8 +179,8 @@ static u64 he_get_##_field(struct hist_entry *he) \
116static int hpp__color_##_type(struct perf_hpp_fmt *fmt __maybe_unused, \ 179static int hpp__color_##_type(struct perf_hpp_fmt *fmt __maybe_unused, \
117 struct perf_hpp *hpp, struct hist_entry *he) \ 180 struct perf_hpp *hpp, struct hist_entry *he) \
118{ \ 181{ \
119 return __hpp__fmt(hpp, he, he_get_##_field, " %6.2f%%", \ 182 return __hpp__fmt(hpp, he, he_get_##_field, NULL, " %6.2f%%", \
120 percent_color_snprintf, true); \ 183 hpp_color_scnprintf, true); \
121} 184}
122 185
123#define __HPP_ENTRY_PERCENT_FN(_type, _field) \ 186#define __HPP_ENTRY_PERCENT_FN(_type, _field) \
@@ -125,8 +188,8 @@ static int hpp__entry_##_type(struct perf_hpp_fmt *_fmt __maybe_unused, \
125 struct perf_hpp *hpp, struct hist_entry *he) \ 188 struct perf_hpp *hpp, struct hist_entry *he) \
126{ \ 189{ \
127 const char *fmt = symbol_conf.field_sep ? " %.2f" : " %6.2f%%"; \ 190 const char *fmt = symbol_conf.field_sep ? " %.2f" : " %6.2f%%"; \
128 return __hpp__fmt(hpp, he, he_get_##_field, fmt, \ 191 return __hpp__fmt(hpp, he, he_get_##_field, NULL, fmt, \
129 scnprintf, true); \ 192 hpp_entry_scnprintf, true); \
130} 193}
131 194
132#define __HPP_ENTRY_RAW_FN(_type, _field) \ 195#define __HPP_ENTRY_RAW_FN(_type, _field) \
@@ -139,7 +202,8 @@ static int hpp__entry_##_type(struct perf_hpp_fmt *_fmt __maybe_unused, \
139 struct perf_hpp *hpp, struct hist_entry *he) \ 202 struct perf_hpp *hpp, struct hist_entry *he) \
140{ \ 203{ \
141 const char *fmt = symbol_conf.field_sep ? " %"PRIu64 : " %11"PRIu64; \ 204 const char *fmt = symbol_conf.field_sep ? " %"PRIu64 : " %11"PRIu64; \
142 return __hpp__fmt(hpp, he, he_get_raw_##_field, fmt, scnprintf, false); \ 205 return __hpp__fmt(hpp, he, he_get_raw_##_field, NULL, fmt, \
206 hpp_entry_scnprintf, false); \
143} 207}
144 208
145#define HPP_PERCENT_FNS(_type, _str, _field, _min_width, _unit_width) \ 209#define HPP_PERCENT_FNS(_type, _str, _field, _min_width, _unit_width) \
@@ -263,15 +327,13 @@ unsigned int hists__sort_list_width(struct hists *hists)
263 struct perf_hpp_fmt *fmt; 327 struct perf_hpp_fmt *fmt;
264 struct sort_entry *se; 328 struct sort_entry *se;
265 int i = 0, ret = 0; 329 int i = 0, ret = 0;
266 struct perf_hpp dummy_hpp = { 330 struct perf_hpp dummy_hpp;
267 .ptr = hists_to_evsel(hists),
268 };
269 331
270 perf_hpp__for_each_format(fmt) { 332 perf_hpp__for_each_format(fmt) {
271 if (i) 333 if (i)
272 ret += 2; 334 ret += 2;
273 335
274 ret += fmt->width(fmt, &dummy_hpp); 336 ret += fmt->width(fmt, &dummy_hpp, hists_to_evsel(hists));
275 } 337 }
276 338
277 list_for_each_entry(se, &hist_entry__sort_list, list) 339 list_for_each_entry(se, &hist_entry__sort_list, list)
diff --git a/tools/perf/ui/stdio/hist.c b/tools/perf/ui/stdio/hist.c
index 831fbb77d1ff..d59893edf031 100644
--- a/tools/perf/ui/stdio/hist.c
+++ b/tools/perf/ui/stdio/hist.c
@@ -306,12 +306,6 @@ static size_t hist_entry__callchain_fprintf(struct hist_entry *he,
306 return hist_entry_callchain__fprintf(he, total_period, left_margin, fp); 306 return hist_entry_callchain__fprintf(he, total_period, left_margin, fp);
307} 307}
308 308
309static inline void advance_hpp(struct perf_hpp *hpp, int inc)
310{
311 hpp->buf += inc;
312 hpp->size -= inc;
313}
314
315static int hist_entry__period_snprintf(struct perf_hpp *hpp, 309static int hist_entry__period_snprintf(struct perf_hpp *hpp,
316 struct hist_entry *he) 310 struct hist_entry *he)
317{ 311{
@@ -385,7 +379,6 @@ size_t hists__fprintf(struct hists *hists, bool show_header, int max_rows,
385 struct perf_hpp dummy_hpp = { 379 struct perf_hpp dummy_hpp = {
386 .buf = bf, 380 .buf = bf,
387 .size = sizeof(bf), 381 .size = sizeof(bf),
388 .ptr = hists_to_evsel(hists),
389 }; 382 };
390 bool first = true; 383 bool first = true;
391 size_t linesz; 384 size_t linesz;
@@ -404,7 +397,7 @@ size_t hists__fprintf(struct hists *hists, bool show_header, int max_rows,
404 else 397 else
405 first = false; 398 first = false;
406 399
407 fmt->header(fmt, &dummy_hpp); 400 fmt->header(fmt, &dummy_hpp, hists_to_evsel(hists));
408 fprintf(fp, "%s", bf); 401 fprintf(fp, "%s", bf);
409 } 402 }
410 403
@@ -449,7 +442,7 @@ size_t hists__fprintf(struct hists *hists, bool show_header, int max_rows,
449 else 442 else
450 first = false; 443 first = false;
451 444
452 width = fmt->width(fmt, &dummy_hpp); 445 width = fmt->width(fmt, &dummy_hpp, hists_to_evsel(hists));
453 for (i = 0; i < width; i++) 446 for (i = 0; i < width; i++)
454 fprintf(fp, "."); 447 fprintf(fp, ".");
455 } 448 }
diff --git a/tools/perf/util/annotate.c b/tools/perf/util/annotate.c
index 3aa555ff9d89..809b4c50beae 100644
--- a/tools/perf/util/annotate.c
+++ b/tools/perf/util/annotate.c
@@ -1236,6 +1236,7 @@ int symbol__annotate_printf(struct symbol *sym, struct map *map,
1236 struct dso *dso = map->dso; 1236 struct dso *dso = map->dso;
1237 char *filename; 1237 char *filename;
1238 const char *d_filename; 1238 const char *d_filename;
1239 const char *evsel_name = perf_evsel__name(evsel);
1239 struct annotation *notes = symbol__annotation(sym); 1240 struct annotation *notes = symbol__annotation(sym);
1240 struct disasm_line *pos, *queue = NULL; 1241 struct disasm_line *pos, *queue = NULL;
1241 u64 start = map__rip_2objdump(map, sym->start); 1242 u64 start = map__rip_2objdump(map, sym->start);
@@ -1243,7 +1244,7 @@ int symbol__annotate_printf(struct symbol *sym, struct map *map,
1243 int more = 0; 1244 int more = 0;
1244 u64 len; 1245 u64 len;
1245 int width = 8; 1246 int width = 8;
1246 int namelen; 1247 int namelen, evsel_name_len, graph_dotted_len;
1247 1248
1248 filename = strdup(dso->long_name); 1249 filename = strdup(dso->long_name);
1249 if (!filename) 1250 if (!filename)
@@ -1256,14 +1257,17 @@ int symbol__annotate_printf(struct symbol *sym, struct map *map,
1256 1257
1257 len = symbol__size(sym); 1258 len = symbol__size(sym);
1258 namelen = strlen(d_filename); 1259 namelen = strlen(d_filename);
1260 evsel_name_len = strlen(evsel_name);
1259 1261
1260 if (perf_evsel__is_group_event(evsel)) 1262 if (perf_evsel__is_group_event(evsel))
1261 width *= evsel->nr_members; 1263 width *= evsel->nr_members;
1262 1264
1263 printf(" %-*.*s| Source code & Disassembly of %s\n", 1265 printf(" %-*.*s| Source code & Disassembly of %s for %s\n",
1264 width, width, "Percent", d_filename); 1266 width, width, "Percent", d_filename, evsel_name);
1265 printf("-%-*.*s-------------------------------------\n", 1267
1266 width+namelen, width+namelen, graph_dotted_line); 1268 graph_dotted_len = width + namelen + evsel_name_len;
1269 printf("-%-*.*s-----------------------------------------\n",
1270 graph_dotted_len, graph_dotted_len, graph_dotted_line);
1267 1271
1268 if (verbose) 1272 if (verbose)
1269 symbol__annotate_hits(sym, evsel); 1273 symbol__annotate_hits(sym, evsel);
diff --git a/tools/perf/util/cpumap.c b/tools/perf/util/cpumap.c
index a9b48c42e81e..7fe4994eeb63 100644
--- a/tools/perf/util/cpumap.c
+++ b/tools/perf/util/cpumap.c
@@ -1,5 +1,5 @@
1#include "util.h" 1#include "util.h"
2#include "fs.h" 2#include <api/fs/fs.h>
3#include "../perf.h" 3#include "../perf.h"
4#include "cpumap.h" 4#include "cpumap.h"
5#include <assert.h> 5#include <assert.h>
diff --git a/tools/perf/util/dso.c b/tools/perf/util/dso.c
index 4045d086d9d9..64453d63b971 100644
--- a/tools/perf/util/dso.c
+++ b/tools/perf/util/dso.c
@@ -45,8 +45,8 @@ int dso__read_binary_type_filename(const struct dso *dso,
45 debuglink--; 45 debuglink--;
46 if (*debuglink == '/') 46 if (*debuglink == '/')
47 debuglink++; 47 debuglink++;
48 filename__read_debuglink(dso->long_name, debuglink, 48 ret = filename__read_debuglink(dso->long_name, debuglink,
49 size - (debuglink - filename)); 49 size - (debuglink - filename));
50 } 50 }
51 break; 51 break;
52 case DSO_BINARY_TYPE__BUILD_ID_CACHE: 52 case DSO_BINARY_TYPE__BUILD_ID_CACHE:
diff --git a/tools/perf/util/dso.h b/tools/perf/util/dso.h
index cd7d6f078cdd..ab06f1c03655 100644
--- a/tools/perf/util/dso.h
+++ b/tools/perf/util/dso.h
@@ -102,6 +102,16 @@ struct dso {
102 char name[0]; 102 char name[0];
103}; 103};
104 104
105/* dso__for_each_symbol - iterate over the symbols of given type
106 *
107 * @dso: the 'struct dso *' in which symbols itereated
108 * @pos: the 'struct symbol *' to use as a loop cursor
109 * @n: the 'struct rb_node *' to use as a temporary storage
110 * @type: the 'enum map_type' type of symbols
111 */
112#define dso__for_each_symbol(dso, pos, n, type) \
113 symbols__for_each_entry(&(dso)->symbols[(type)], pos, n)
114
105static inline void dso__set_loaded(struct dso *dso, enum map_type type) 115static inline void dso__set_loaded(struct dso *dso, enum map_type type)
106{ 116{
107 dso->loaded |= (1 << type); 117 dso->loaded |= (1 << type);
diff --git a/tools/perf/util/event.c b/tools/perf/util/event.c
index b0f3ca850e9e..9d12aa6dd485 100644
--- a/tools/perf/util/event.c
+++ b/tools/perf/util/event.c
@@ -1,6 +1,7 @@
1#include <linux/types.h> 1#include <linux/types.h>
2#include "event.h" 2#include "event.h"
3#include "debug.h" 3#include "debug.h"
4#include "hist.h"
4#include "machine.h" 5#include "machine.h"
5#include "sort.h" 6#include "sort.h"
6#include "string.h" 7#include "string.h"
@@ -94,14 +95,10 @@ static pid_t perf_event__get_comm_tgid(pid_t pid, char *comm, size_t len)
94 95
95static pid_t perf_event__synthesize_comm(struct perf_tool *tool, 96static pid_t perf_event__synthesize_comm(struct perf_tool *tool,
96 union perf_event *event, pid_t pid, 97 union perf_event *event, pid_t pid,
97 int full,
98 perf_event__handler_t process, 98 perf_event__handler_t process,
99 struct machine *machine) 99 struct machine *machine)
100{ 100{
101 char filename[PATH_MAX];
102 size_t size; 101 size_t size;
103 DIR *tasks;
104 struct dirent dirent, *next;
105 pid_t tgid; 102 pid_t tgid;
106 103
107 memset(&event->comm, 0, sizeof(event->comm)); 104 memset(&event->comm, 0, sizeof(event->comm));
@@ -124,55 +121,35 @@ static pid_t perf_event__synthesize_comm(struct perf_tool *tool,
124 event->comm.header.size = (sizeof(event->comm) - 121 event->comm.header.size = (sizeof(event->comm) -
125 (sizeof(event->comm.comm) - size) + 122 (sizeof(event->comm.comm) - size) +
126 machine->id_hdr_size); 123 machine->id_hdr_size);
127 if (!full) { 124 event->comm.tid = pid;
128 event->comm.tid = pid;
129 125
130 if (process(tool, event, &synth_sample, machine) != 0) 126 if (process(tool, event, &synth_sample, machine) != 0)
131 return -1; 127 return -1;
132
133 goto out;
134 }
135
136 if (machine__is_default_guest(machine))
137 return 0;
138
139 snprintf(filename, sizeof(filename), "%s/proc/%d/task",
140 machine->root_dir, pid);
141
142 tasks = opendir(filename);
143 if (tasks == NULL) {
144 pr_debug("couldn't open %s\n", filename);
145 return 0;
146 }
147 128
148 while (!readdir_r(tasks, &dirent, &next) && next) { 129out:
149 char *end; 130 return tgid;
150 pid = strtol(dirent.d_name, &end, 10); 131}
151 if (*end)
152 continue;
153 132
154 /* already have tgid; jut want to update the comm */ 133static int perf_event__synthesize_fork(struct perf_tool *tool,
155 (void) perf_event__get_comm_tgid(pid, event->comm.comm, 134 union perf_event *event, pid_t pid,
156 sizeof(event->comm.comm)); 135 pid_t tgid, perf_event__handler_t process,
136 struct machine *machine)
137{
138 memset(&event->fork, 0, sizeof(event->fork) + machine->id_hdr_size);
157 139
158 size = strlen(event->comm.comm) + 1; 140 /* this is really a clone event but we use fork to synthesize it */
159 size = PERF_ALIGN(size, sizeof(u64)); 141 event->fork.ppid = tgid;
160 memset(event->comm.comm + size, 0, machine->id_hdr_size); 142 event->fork.ptid = tgid;
161 event->comm.header.size = (sizeof(event->comm) - 143 event->fork.pid = tgid;
162 (sizeof(event->comm.comm) - size) + 144 event->fork.tid = pid;
163 machine->id_hdr_size); 145 event->fork.header.type = PERF_RECORD_FORK;
164 146
165 event->comm.tid = pid; 147 event->fork.header.size = (sizeof(event->fork) + machine->id_hdr_size);
166 148
167 if (process(tool, event, &synth_sample, machine) != 0) { 149 if (process(tool, event, &synth_sample, machine) != 0)
168 tgid = -1; 150 return -1;
169 break;
170 }
171 }
172 151
173 closedir(tasks); 152 return 0;
174out:
175 return tgid;
176} 153}
177 154
178int perf_event__synthesize_mmap_events(struct perf_tool *tool, 155int perf_event__synthesize_mmap_events(struct perf_tool *tool,
@@ -324,17 +301,71 @@ int perf_event__synthesize_modules(struct perf_tool *tool,
324 301
325static int __event__synthesize_thread(union perf_event *comm_event, 302static int __event__synthesize_thread(union perf_event *comm_event,
326 union perf_event *mmap_event, 303 union perf_event *mmap_event,
304 union perf_event *fork_event,
327 pid_t pid, int full, 305 pid_t pid, int full,
328 perf_event__handler_t process, 306 perf_event__handler_t process,
329 struct perf_tool *tool, 307 struct perf_tool *tool,
330 struct machine *machine, bool mmap_data) 308 struct machine *machine, bool mmap_data)
331{ 309{
332 pid_t tgid = perf_event__synthesize_comm(tool, comm_event, pid, full, 310 char filename[PATH_MAX];
311 DIR *tasks;
312 struct dirent dirent, *next;
313 pid_t tgid;
314
315 /* special case: only send one comm event using passed in pid */
316 if (!full) {
317 tgid = perf_event__synthesize_comm(tool, comm_event, pid,
318 process, machine);
319
320 if (tgid == -1)
321 return -1;
322
323 return perf_event__synthesize_mmap_events(tool, mmap_event, pid, tgid,
324 process, machine, mmap_data);
325 }
326
327 if (machine__is_default_guest(machine))
328 return 0;
329
330 snprintf(filename, sizeof(filename), "%s/proc/%d/task",
331 machine->root_dir, pid);
332
333 tasks = opendir(filename);
334 if (tasks == NULL) {
335 pr_debug("couldn't open %s\n", filename);
336 return 0;
337 }
338
339 while (!readdir_r(tasks, &dirent, &next) && next) {
340 char *end;
341 int rc = 0;
342 pid_t _pid;
343
344 _pid = strtol(dirent.d_name, &end, 10);
345 if (*end)
346 continue;
347
348 tgid = perf_event__synthesize_comm(tool, comm_event, _pid,
349 process, machine);
350 if (tgid == -1)
351 return -1;
352
353 if (_pid == pid) {
354 /* process the parent's maps too */
355 rc = perf_event__synthesize_mmap_events(tool, mmap_event, pid, tgid,
356 process, machine, mmap_data);
357 } else {
358 /* only fork the tid's map, to save time */
359 rc = perf_event__synthesize_fork(tool, fork_event, _pid, tgid,
333 process, machine); 360 process, machine);
334 if (tgid == -1) 361 }
335 return -1; 362
336 return perf_event__synthesize_mmap_events(tool, mmap_event, pid, tgid, 363 if (rc)
337 process, machine, mmap_data); 364 return rc;
365 }
366
367 closedir(tasks);
368 return 0;
338} 369}
339 370
340int perf_event__synthesize_thread_map(struct perf_tool *tool, 371int perf_event__synthesize_thread_map(struct perf_tool *tool,
@@ -343,7 +374,7 @@ int perf_event__synthesize_thread_map(struct perf_tool *tool,
343 struct machine *machine, 374 struct machine *machine,
344 bool mmap_data) 375 bool mmap_data)
345{ 376{
346 union perf_event *comm_event, *mmap_event; 377 union perf_event *comm_event, *mmap_event, *fork_event;
347 int err = -1, thread, j; 378 int err = -1, thread, j;
348 379
349 comm_event = malloc(sizeof(comm_event->comm) + machine->id_hdr_size); 380 comm_event = malloc(sizeof(comm_event->comm) + machine->id_hdr_size);
@@ -354,9 +385,14 @@ int perf_event__synthesize_thread_map(struct perf_tool *tool,
354 if (mmap_event == NULL) 385 if (mmap_event == NULL)
355 goto out_free_comm; 386 goto out_free_comm;
356 387
388 fork_event = malloc(sizeof(fork_event->fork) + machine->id_hdr_size);
389 if (fork_event == NULL)
390 goto out_free_mmap;
391
357 err = 0; 392 err = 0;
358 for (thread = 0; thread < threads->nr; ++thread) { 393 for (thread = 0; thread < threads->nr; ++thread) {
359 if (__event__synthesize_thread(comm_event, mmap_event, 394 if (__event__synthesize_thread(comm_event, mmap_event,
395 fork_event,
360 threads->map[thread], 0, 396 threads->map[thread], 0,
361 process, tool, machine, 397 process, tool, machine,
362 mmap_data)) { 398 mmap_data)) {
@@ -382,6 +418,7 @@ int perf_event__synthesize_thread_map(struct perf_tool *tool,
382 /* if not, generate events for it */ 418 /* if not, generate events for it */
383 if (need_leader && 419 if (need_leader &&
384 __event__synthesize_thread(comm_event, mmap_event, 420 __event__synthesize_thread(comm_event, mmap_event,
421 fork_event,
385 comm_event->comm.pid, 0, 422 comm_event->comm.pid, 0,
386 process, tool, machine, 423 process, tool, machine,
387 mmap_data)) { 424 mmap_data)) {
@@ -390,6 +427,8 @@ int perf_event__synthesize_thread_map(struct perf_tool *tool,
390 } 427 }
391 } 428 }
392 } 429 }
430 free(fork_event);
431out_free_mmap:
393 free(mmap_event); 432 free(mmap_event);
394out_free_comm: 433out_free_comm:
395 free(comm_event); 434 free(comm_event);
@@ -404,9 +443,12 @@ int perf_event__synthesize_threads(struct perf_tool *tool,
404 DIR *proc; 443 DIR *proc;
405 char proc_path[PATH_MAX]; 444 char proc_path[PATH_MAX];
406 struct dirent dirent, *next; 445 struct dirent dirent, *next;
407 union perf_event *comm_event, *mmap_event; 446 union perf_event *comm_event, *mmap_event, *fork_event;
408 int err = -1; 447 int err = -1;
409 448
449 if (machine__is_default_guest(machine))
450 return 0;
451
410 comm_event = malloc(sizeof(comm_event->comm) + machine->id_hdr_size); 452 comm_event = malloc(sizeof(comm_event->comm) + machine->id_hdr_size);
411 if (comm_event == NULL) 453 if (comm_event == NULL)
412 goto out; 454 goto out;
@@ -415,14 +457,15 @@ int perf_event__synthesize_threads(struct perf_tool *tool,
415 if (mmap_event == NULL) 457 if (mmap_event == NULL)
416 goto out_free_comm; 458 goto out_free_comm;
417 459
418 if (machine__is_default_guest(machine)) 460 fork_event = malloc(sizeof(fork_event->fork) + machine->id_hdr_size);
419 return 0; 461 if (fork_event == NULL)
462 goto out_free_mmap;
420 463
421 snprintf(proc_path, sizeof(proc_path), "%s/proc", machine->root_dir); 464 snprintf(proc_path, sizeof(proc_path), "%s/proc", machine->root_dir);
422 proc = opendir(proc_path); 465 proc = opendir(proc_path);
423 466
424 if (proc == NULL) 467 if (proc == NULL)
425 goto out_free_mmap; 468 goto out_free_fork;
426 469
427 while (!readdir_r(proc, &dirent, &next) && next) { 470 while (!readdir_r(proc, &dirent, &next) && next) {
428 char *end; 471 char *end;
@@ -434,12 +477,14 @@ int perf_event__synthesize_threads(struct perf_tool *tool,
434 * We may race with exiting thread, so don't stop just because 477 * We may race with exiting thread, so don't stop just because
435 * one thread couldn't be synthesized. 478 * one thread couldn't be synthesized.
436 */ 479 */
437 __event__synthesize_thread(comm_event, mmap_event, pid, 1, 480 __event__synthesize_thread(comm_event, mmap_event, fork_event, pid,
438 process, tool, machine, mmap_data); 481 1, process, tool, machine, mmap_data);
439 } 482 }
440 483
441 err = 0; 484 err = 0;
442 closedir(proc); 485 closedir(proc);
486out_free_fork:
487 free(fork_event);
443out_free_mmap: 488out_free_mmap:
444 free(mmap_event); 489 free(mmap_event);
445out_free_comm: 490out_free_comm:
@@ -661,7 +706,7 @@ void thread__find_addr_map(struct thread *thread,
661 al->thread = thread; 706 al->thread = thread;
662 al->addr = addr; 707 al->addr = addr;
663 al->cpumode = cpumode; 708 al->cpumode = cpumode;
664 al->filtered = false; 709 al->filtered = 0;
665 710
666 if (machine == NULL) { 711 if (machine == NULL) {
667 al->map = NULL; 712 al->map = NULL;
@@ -687,11 +732,11 @@ void thread__find_addr_map(struct thread *thread,
687 if ((cpumode == PERF_RECORD_MISC_GUEST_USER || 732 if ((cpumode == PERF_RECORD_MISC_GUEST_USER ||
688 cpumode == PERF_RECORD_MISC_GUEST_KERNEL) && 733 cpumode == PERF_RECORD_MISC_GUEST_KERNEL) &&
689 !perf_guest) 734 !perf_guest)
690 al->filtered = true; 735 al->filtered |= (1 << HIST_FILTER__GUEST);
691 if ((cpumode == PERF_RECORD_MISC_USER || 736 if ((cpumode == PERF_RECORD_MISC_USER ||
692 cpumode == PERF_RECORD_MISC_KERNEL) && 737 cpumode == PERF_RECORD_MISC_KERNEL) &&
693 !perf_host) 738 !perf_host)
694 al->filtered = true; 739 al->filtered |= (1 << HIST_FILTER__HOST);
695 740
696 return; 741 return;
697 } 742 }
@@ -748,9 +793,6 @@ int perf_event__preprocess_sample(const union perf_event *event,
748 if (thread == NULL) 793 if (thread == NULL)
749 return -1; 794 return -1;
750 795
751 if (thread__is_filtered(thread))
752 goto out_filtered;
753
754 dump_printf(" ... thread: %s:%d\n", thread__comm_str(thread), thread->tid); 796 dump_printf(" ... thread: %s:%d\n", thread__comm_str(thread), thread->tid);
755 /* 797 /*
756 * Have we already created the kernel maps for this machine? 798 * Have we already created the kernel maps for this machine?
@@ -768,6 +810,10 @@ int perf_event__preprocess_sample(const union perf_event *event,
768 dump_printf(" ...... dso: %s\n", 810 dump_printf(" ...... dso: %s\n",
769 al->map ? al->map->dso->long_name : 811 al->map ? al->map->dso->long_name :
770 al->level == 'H' ? "[hypervisor]" : "<not found>"); 812 al->level == 'H' ? "[hypervisor]" : "<not found>");
813
814 if (thread__is_filtered(thread))
815 al->filtered |= (1 << HIST_FILTER__THREAD);
816
771 al->sym = NULL; 817 al->sym = NULL;
772 al->cpu = sample->cpu; 818 al->cpu = sample->cpu;
773 819
@@ -779,8 +825,9 @@ int perf_event__preprocess_sample(const union perf_event *event,
779 dso->short_name) || 825 dso->short_name) ||
780 (dso->short_name != dso->long_name && 826 (dso->short_name != dso->long_name &&
781 strlist__has_entry(symbol_conf.dso_list, 827 strlist__has_entry(symbol_conf.dso_list,
782 dso->long_name))))) 828 dso->long_name))))) {
783 goto out_filtered; 829 al->filtered |= (1 << HIST_FILTER__DSO);
830 }
784 831
785 al->sym = map__find_symbol(al->map, al->addr, 832 al->sym = map__find_symbol(al->map, al->addr,
786 machine->symbol_filter); 833 machine->symbol_filter);
@@ -788,12 +835,9 @@ int perf_event__preprocess_sample(const union perf_event *event,
788 835
789 if (symbol_conf.sym_list && 836 if (symbol_conf.sym_list &&
790 (!al->sym || !strlist__has_entry(symbol_conf.sym_list, 837 (!al->sym || !strlist__has_entry(symbol_conf.sym_list,
791 al->sym->name))) 838 al->sym->name))) {
792 goto out_filtered; 839 al->filtered |= (1 << HIST_FILTER__SYMBOL);
793 840 }
794 return 0;
795 841
796out_filtered:
797 al->filtered = true;
798 return 0; 842 return 0;
799} 843}
diff --git a/tools/perf/util/event.h b/tools/perf/util/event.h
index 851fa06f4a42..38457d447a13 100644
--- a/tools/perf/util/event.h
+++ b/tools/perf/util/event.h
@@ -85,6 +85,7 @@ struct sample_event {
85 85
86struct regs_dump { 86struct regs_dump {
87 u64 abi; 87 u64 abi;
88 u64 mask;
88 u64 *regs; 89 u64 *regs;
89}; 90};
90 91
@@ -259,9 +260,9 @@ int perf_event__preprocess_sample(const union perf_event *event,
259const char *perf_event__name(unsigned int id); 260const char *perf_event__name(unsigned int id);
260 261
261size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type, 262size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
262 u64 sample_regs_user, u64 read_format); 263 u64 read_format);
263int perf_event__synthesize_sample(union perf_event *event, u64 type, 264int perf_event__synthesize_sample(union perf_event *event, u64 type,
264 u64 sample_regs_user, u64 read_format, 265 u64 read_format,
265 const struct perf_sample *sample, 266 const struct perf_sample *sample,
266 bool swapped); 267 bool swapped);
267 268
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 55407c594b87..5c28d82b76c4 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -500,6 +500,34 @@ int perf_evsel__group_desc(struct perf_evsel *evsel, char *buf, size_t size)
500 return ret; 500 return ret;
501} 501}
502 502
503static void
504perf_evsel__config_callgraph(struct perf_evsel *evsel,
505 struct record_opts *opts)
506{
507 bool function = perf_evsel__is_function_event(evsel);
508 struct perf_event_attr *attr = &evsel->attr;
509
510 perf_evsel__set_sample_bit(evsel, CALLCHAIN);
511
512 if (opts->call_graph == CALLCHAIN_DWARF) {
513 if (!function) {
514 perf_evsel__set_sample_bit(evsel, REGS_USER);
515 perf_evsel__set_sample_bit(evsel, STACK_USER);
516 attr->sample_regs_user = PERF_REGS_MASK;
517 attr->sample_stack_user = opts->stack_dump_size;
518 attr->exclude_callchain_user = 1;
519 } else {
520 pr_info("Cannot use DWARF unwind for function trace event,"
521 " falling back to framepointers.\n");
522 }
523 }
524
525 if (function) {
526 pr_info("Disabling user space callchains for function trace event.\n");
527 attr->exclude_callchain_user = 1;
528 }
529}
530
503/* 531/*
504 * The enable_on_exec/disabled value strategy: 532 * The enable_on_exec/disabled value strategy:
505 * 533 *
@@ -595,17 +623,8 @@ void perf_evsel__config(struct perf_evsel *evsel, struct record_opts *opts)
595 attr->mmap_data = track; 623 attr->mmap_data = track;
596 } 624 }
597 625
598 if (opts->call_graph) { 626 if (opts->call_graph_enabled)
599 perf_evsel__set_sample_bit(evsel, CALLCHAIN); 627 perf_evsel__config_callgraph(evsel, opts);
600
601 if (opts->call_graph == CALLCHAIN_DWARF) {
602 perf_evsel__set_sample_bit(evsel, REGS_USER);
603 perf_evsel__set_sample_bit(evsel, STACK_USER);
604 attr->sample_regs_user = PERF_REGS_MASK;
605 attr->sample_stack_user = opts->stack_dump_size;
606 attr->exclude_callchain_user = 1;
607 }
608 }
609 628
610 if (target__has_cpu(&opts->target)) 629 if (target__has_cpu(&opts->target))
611 perf_evsel__set_sample_bit(evsel, CPU); 630 perf_evsel__set_sample_bit(evsel, CPU);
@@ -1004,7 +1023,7 @@ retry_sample_id:
1004 1023
1005 group_fd = get_group_fd(evsel, cpu, thread); 1024 group_fd = get_group_fd(evsel, cpu, thread);
1006retry_open: 1025retry_open:
1007 pr_debug2("perf_event_open: pid %d cpu %d group_fd %d flags %#lx\n", 1026 pr_debug2("sys_perf_event_open: pid %d cpu %d group_fd %d flags %#lx\n",
1008 pid, cpus->map[cpu], group_fd, flags); 1027 pid, cpus->map[cpu], group_fd, flags);
1009 1028
1010 FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr, 1029 FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr,
@@ -1013,7 +1032,7 @@ retry_open:
1013 group_fd, flags); 1032 group_fd, flags);
1014 if (FD(evsel, cpu, thread) < 0) { 1033 if (FD(evsel, cpu, thread) < 0) {
1015 err = -errno; 1034 err = -errno;
1016 pr_debug2("perf_event_open failed, error %d\n", 1035 pr_debug2("sys_perf_event_open failed, error %d\n",
1017 err); 1036 err);
1018 goto try_fallback; 1037 goto try_fallback;
1019 } 1038 }
@@ -1220,7 +1239,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
1220 memset(data, 0, sizeof(*data)); 1239 memset(data, 0, sizeof(*data));
1221 data->cpu = data->pid = data->tid = -1; 1240 data->cpu = data->pid = data->tid = -1;
1222 data->stream_id = data->id = data->time = -1ULL; 1241 data->stream_id = data->id = data->time = -1ULL;
1223 data->period = 1; 1242 data->period = evsel->attr.sample_period;
1224 data->weight = 0; 1243 data->weight = 0;
1225 1244
1226 if (event->header.type != PERF_RECORD_SAMPLE) { 1245 if (event->header.type != PERF_RECORD_SAMPLE) {
@@ -1396,10 +1415,11 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
1396 array++; 1415 array++;
1397 1416
1398 if (data->user_regs.abi) { 1417 if (data->user_regs.abi) {
1399 u64 regs_user = evsel->attr.sample_regs_user; 1418 u64 mask = evsel->attr.sample_regs_user;
1400 1419
1401 sz = hweight_long(regs_user) * sizeof(u64); 1420 sz = hweight_long(mask) * sizeof(u64);
1402 OVERFLOW_CHECK(array, sz, max_size); 1421 OVERFLOW_CHECK(array, sz, max_size);
1422 data->user_regs.mask = mask;
1403 data->user_regs.regs = (u64 *)array; 1423 data->user_regs.regs = (u64 *)array;
1404 array = (void *)array + sz; 1424 array = (void *)array + sz;
1405 } 1425 }
@@ -1451,7 +1471,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
1451} 1471}
1452 1472
1453size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type, 1473size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
1454 u64 sample_regs_user, u64 read_format) 1474 u64 read_format)
1455{ 1475{
1456 size_t sz, result = sizeof(struct sample_event); 1476 size_t sz, result = sizeof(struct sample_event);
1457 1477
@@ -1517,7 +1537,7 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
1517 if (type & PERF_SAMPLE_REGS_USER) { 1537 if (type & PERF_SAMPLE_REGS_USER) {
1518 if (sample->user_regs.abi) { 1538 if (sample->user_regs.abi) {
1519 result += sizeof(u64); 1539 result += sizeof(u64);
1520 sz = hweight_long(sample_regs_user) * sizeof(u64); 1540 sz = hweight_long(sample->user_regs.mask) * sizeof(u64);
1521 result += sz; 1541 result += sz;
1522 } else { 1542 } else {
1523 result += sizeof(u64); 1543 result += sizeof(u64);
@@ -1546,7 +1566,7 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
1546} 1566}
1547 1567
1548int perf_event__synthesize_sample(union perf_event *event, u64 type, 1568int perf_event__synthesize_sample(union perf_event *event, u64 type,
1549 u64 sample_regs_user, u64 read_format, 1569 u64 read_format,
1550 const struct perf_sample *sample, 1570 const struct perf_sample *sample,
1551 bool swapped) 1571 bool swapped)
1552{ 1572{
@@ -1687,7 +1707,7 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type,
1687 if (type & PERF_SAMPLE_REGS_USER) { 1707 if (type & PERF_SAMPLE_REGS_USER) {
1688 if (sample->user_regs.abi) { 1708 if (sample->user_regs.abi) {
1689 *array++ = sample->user_regs.abi; 1709 *array++ = sample->user_regs.abi;
1690 sz = hweight_long(sample_regs_user) * sizeof(u64); 1710 sz = hweight_long(sample->user_regs.mask) * sizeof(u64);
1691 memcpy(array, sample->user_regs.regs, sz); 1711 memcpy(array, sample->user_regs.regs, sz);
1692 array = (void *)array + sz; 1712 array = (void *)array + sz;
1693 } else { 1713 } else {
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index f1b325665aae..0c9926cfb292 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -315,6 +315,24 @@ static inline bool perf_evsel__is_group_event(struct perf_evsel *evsel)
315 return perf_evsel__is_group_leader(evsel) && evsel->nr_members > 1; 315 return perf_evsel__is_group_leader(evsel) && evsel->nr_members > 1;
316} 316}
317 317
318/**
319 * perf_evsel__is_function_event - Return whether given evsel is a function
320 * trace event
321 *
322 * @evsel - evsel selector to be tested
323 *
324 * Return %true if event is function trace event
325 */
326static inline bool perf_evsel__is_function_event(struct perf_evsel *evsel)
327{
328#define FUNCTION_EVENT "ftrace:function"
329
330 return evsel->name &&
331 !strncmp(FUNCTION_EVENT, evsel->name, sizeof(FUNCTION_EVENT));
332
333#undef FUNCTION_EVENT
334}
335
318struct perf_attr_details { 336struct perf_attr_details {
319 bool freq; 337 bool freq;
320 bool verbose; 338 bool verbose;
diff --git a/tools/perf/util/fs.h b/tools/perf/util/fs.h
deleted file mode 100644
index 5e09ce1bab0e..000000000000
--- a/tools/perf/util/fs.h
+++ /dev/null
@@ -1,7 +0,0 @@
1#ifndef __PERF_FS
2#define __PERF_FS
3
4const char *sysfs__mountpoint(void);
5const char *procfs__mountpoint(void);
6
7#endif /* __PERF_FS */
diff --git a/tools/perf/util/hist.c b/tools/perf/util/hist.c
index e4e6249b87d4..f38590d7561b 100644
--- a/tools/perf/util/hist.c
+++ b/tools/perf/util/hist.c
@@ -13,13 +13,6 @@ static bool hists__filter_entry_by_thread(struct hists *hists,
13static bool hists__filter_entry_by_symbol(struct hists *hists, 13static bool hists__filter_entry_by_symbol(struct hists *hists,
14 struct hist_entry *he); 14 struct hist_entry *he);
15 15
16enum hist_filter {
17 HIST_FILTER__DSO,
18 HIST_FILTER__THREAD,
19 HIST_FILTER__PARENT,
20 HIST_FILTER__SYMBOL,
21};
22
23struct callchain_param callchain_param = { 16struct callchain_param callchain_param = {
24 .mode = CHAIN_GRAPH_REL, 17 .mode = CHAIN_GRAPH_REL,
25 .min_percent = 0.5, 18 .min_percent = 0.5,
@@ -290,7 +283,7 @@ static struct hist_entry *hist_entry__new(struct hist_entry *template)
290 if (he->branch_info) { 283 if (he->branch_info) {
291 /* 284 /*
292 * This branch info is (a part of) allocated from 285 * This branch info is (a part of) allocated from
293 * machine__resolve_bstack() and will be freed after 286 * sample__resolve_bstack() and will be freed after
294 * adding new entries. So we need to save a copy. 287 * adding new entries. So we need to save a copy.
295 */ 288 */
296 he->branch_info = malloc(sizeof(*he->branch_info)); 289 he->branch_info = malloc(sizeof(*he->branch_info));
@@ -369,7 +362,7 @@ static struct hist_entry *add_hist_entry(struct hists *hists,
369 he_stat__add_period(&he->stat, period, weight); 362 he_stat__add_period(&he->stat, period, weight);
370 363
371 /* 364 /*
372 * This mem info was allocated from machine__resolve_mem 365 * This mem info was allocated from sample__resolve_mem
373 * and will not be used anymore. 366 * and will not be used anymore.
374 */ 367 */
375 zfree(&entry->mem_info); 368 zfree(&entry->mem_info);
@@ -429,7 +422,7 @@ struct hist_entry *__hists__add_entry(struct hists *hists,
429 .weight = weight, 422 .weight = weight,
430 }, 423 },
431 .parent = sym_parent, 424 .parent = sym_parent,
432 .filtered = symbol__parent_filter(sym_parent), 425 .filtered = symbol__parent_filter(sym_parent) | al->filtered,
433 .hists = hists, 426 .hists = hists,
434 .branch_info = bi, 427 .branch_info = bi,
435 .mem_info = mi, 428 .mem_info = mi,
diff --git a/tools/perf/util/hist.h b/tools/perf/util/hist.h
index a59743fa3ef7..1f1f513dfe7f 100644
--- a/tools/perf/util/hist.h
+++ b/tools/perf/util/hist.h
@@ -14,6 +14,15 @@ struct hist_entry;
14struct addr_location; 14struct addr_location;
15struct symbol; 15struct symbol;
16 16
17enum hist_filter {
18 HIST_FILTER__DSO,
19 HIST_FILTER__THREAD,
20 HIST_FILTER__PARENT,
21 HIST_FILTER__SYMBOL,
22 HIST_FILTER__GUEST,
23 HIST_FILTER__HOST,
24};
25
17/* 26/*
18 * The kernel collects the number of events it couldn't send in a stretch and 27 * The kernel collects the number of events it couldn't send in a stretch and
19 * when possible sends this number in a PERF_RECORD_LOST event. The number of 28 * when possible sends this number in a PERF_RECORD_LOST event. The number of
@@ -132,8 +141,10 @@ struct perf_hpp {
132}; 141};
133 142
134struct perf_hpp_fmt { 143struct perf_hpp_fmt {
135 int (*header)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp); 144 int (*header)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
136 int (*width)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp); 145 struct perf_evsel *evsel);
146 int (*width)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
147 struct perf_evsel *evsel);
137 int (*color)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, 148 int (*color)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
138 struct hist_entry *he); 149 struct hist_entry *he);
139 int (*entry)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp, 150 int (*entry)(struct perf_hpp_fmt *fmt, struct perf_hpp *hpp,
@@ -166,6 +177,20 @@ void perf_hpp__init(void);
166void perf_hpp__column_register(struct perf_hpp_fmt *format); 177void perf_hpp__column_register(struct perf_hpp_fmt *format);
167void perf_hpp__column_enable(unsigned col); 178void perf_hpp__column_enable(unsigned col);
168 179
180typedef u64 (*hpp_field_fn)(struct hist_entry *he);
181typedef int (*hpp_callback_fn)(struct perf_hpp *hpp, bool front);
182typedef int (*hpp_snprint_fn)(struct perf_hpp *hpp, const char *fmt, ...);
183
184int __hpp__fmt(struct perf_hpp *hpp, struct hist_entry *he,
185 hpp_field_fn get_field, hpp_callback_fn callback,
186 const char *fmt, hpp_snprint_fn print_fn, bool fmt_percent);
187
188static inline void advance_hpp(struct perf_hpp *hpp, int inc)
189{
190 hpp->buf += inc;
191 hpp->size -= inc;
192}
193
169static inline size_t perf_hpp__use_color(void) 194static inline size_t perf_hpp__use_color(void)
170{ 195{
171 return !symbol_conf.field_sep; 196 return !symbol_conf.field_sep;
diff --git a/tools/perf/util/include/linux/hash.h b/tools/perf/util/include/linux/hash.h
deleted file mode 100644
index 201f57397997..000000000000
--- a/tools/perf/util/include/linux/hash.h
+++ /dev/null
@@ -1,5 +0,0 @@
1#include "../../../../include/linux/hash.h"
2
3#ifndef PERF_HASH_H
4#define PERF_HASH_H
5#endif
diff --git a/tools/perf/util/include/linux/kernel.h b/tools/perf/util/include/linux/kernel.h
index d8c927c868ee..9844c31b7c2b 100644
--- a/tools/perf/util/include/linux/kernel.h
+++ b/tools/perf/util/include/linux/kernel.h
@@ -94,12 +94,6 @@ static inline int scnprintf(char * buf, size_t size, const char * fmt, ...)
94 return (i >= ssize) ? (ssize - 1) : i; 94 return (i >= ssize) ? (ssize - 1) : i;
95} 95}
96 96
97static inline unsigned long
98simple_strtoul(const char *nptr, char **endptr, int base)
99{
100 return strtoul(nptr, endptr, base);
101}
102
103int eprintf(int level, 97int eprintf(int level,
104 const char *fmt, ...) __attribute__((format(printf, 2, 3))); 98 const char *fmt, ...) __attribute__((format(printf, 2, 3)));
105 99
diff --git a/tools/perf/util/include/linux/list.h b/tools/perf/util/include/linux/list.h
index 1d928a0ce997..bfe0a2afd0d2 100644
--- a/tools/perf/util/include/linux/list.h
+++ b/tools/perf/util/include/linux/list.h
@@ -1,5 +1,4 @@
1#include <linux/kernel.h> 1#include <linux/kernel.h>
2#include <linux/prefetch.h>
3 2
4#include "../../../../include/linux/list.h" 3#include "../../../../include/linux/list.h"
5 4
diff --git a/tools/perf/util/include/linux/prefetch.h b/tools/perf/util/include/linux/prefetch.h
deleted file mode 100644
index 7841e485d8c3..000000000000
--- a/tools/perf/util/include/linux/prefetch.h
+++ /dev/null
@@ -1,6 +0,0 @@
1#ifndef PERF_LINUX_PREFETCH_H
2#define PERF_LINUX_PREFETCH_H
3
4static inline void prefetch(void *a __attribute__((unused))) { }
5
6#endif
diff --git a/tools/perf/util/machine.c b/tools/perf/util/machine.c
index 620a1983b76b..a53cd0b8c151 100644
--- a/tools/perf/util/machine.c
+++ b/tools/perf/util/machine.c
@@ -327,9 +327,10 @@ struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
327 return __machine__findnew_thread(machine, pid, tid, true); 327 return __machine__findnew_thread(machine, pid, tid, true);
328} 328}
329 329
330struct thread *machine__find_thread(struct machine *machine, pid_t tid) 330struct thread *machine__find_thread(struct machine *machine, pid_t pid,
331 pid_t tid)
331{ 332{
332 return __machine__findnew_thread(machine, 0, tid, false); 333 return __machine__findnew_thread(machine, pid, tid, false);
333} 334}
334 335
335int machine__process_comm_event(struct machine *machine, union perf_event *event, 336int machine__process_comm_event(struct machine *machine, union perf_event *event,
@@ -1026,7 +1027,7 @@ int machine__process_mmap2_event(struct machine *machine,
1026 } 1027 }
1027 1028
1028 thread = machine__findnew_thread(machine, event->mmap2.pid, 1029 thread = machine__findnew_thread(machine, event->mmap2.pid,
1029 event->mmap2.pid); 1030 event->mmap2.tid);
1030 if (thread == NULL) 1031 if (thread == NULL)
1031 goto out_problem; 1032 goto out_problem;
1032 1033
@@ -1074,7 +1075,7 @@ int machine__process_mmap_event(struct machine *machine, union perf_event *event
1074 } 1075 }
1075 1076
1076 thread = machine__findnew_thread(machine, event->mmap.pid, 1077 thread = machine__findnew_thread(machine, event->mmap.pid,
1077 event->mmap.pid); 1078 event->mmap.tid);
1078 if (thread == NULL) 1079 if (thread == NULL)
1079 goto out_problem; 1080 goto out_problem;
1080 1081
@@ -1114,7 +1115,9 @@ static void machine__remove_thread(struct machine *machine, struct thread *th)
1114int machine__process_fork_event(struct machine *machine, union perf_event *event, 1115int machine__process_fork_event(struct machine *machine, union perf_event *event,
1115 struct perf_sample *sample) 1116 struct perf_sample *sample)
1116{ 1117{
1117 struct thread *thread = machine__find_thread(machine, event->fork.tid); 1118 struct thread *thread = machine__find_thread(machine,
1119 event->fork.pid,
1120 event->fork.tid);
1118 struct thread *parent = machine__findnew_thread(machine, 1121 struct thread *parent = machine__findnew_thread(machine,
1119 event->fork.ppid, 1122 event->fork.ppid,
1120 event->fork.ptid); 1123 event->fork.ptid);
@@ -1140,7 +1143,9 @@ int machine__process_fork_event(struct machine *machine, union perf_event *event
1140int machine__process_exit_event(struct machine *machine, union perf_event *event, 1143int machine__process_exit_event(struct machine *machine, union perf_event *event,
1141 struct perf_sample *sample __maybe_unused) 1144 struct perf_sample *sample __maybe_unused)
1142{ 1145{
1143 struct thread *thread = machine__find_thread(machine, event->fork.tid); 1146 struct thread *thread = machine__find_thread(machine,
1147 event->fork.pid,
1148 event->fork.tid);
1144 1149
1145 if (dump_trace) 1150 if (dump_trace)
1146 perf_event__fprintf_task(event, stdout); 1151 perf_event__fprintf_task(event, stdout);
@@ -1184,39 +1189,22 @@ static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1184 return 0; 1189 return 0;
1185} 1190}
1186 1191
1187static const u8 cpumodes[] = {
1188 PERF_RECORD_MISC_USER,
1189 PERF_RECORD_MISC_KERNEL,
1190 PERF_RECORD_MISC_GUEST_USER,
1191 PERF_RECORD_MISC_GUEST_KERNEL
1192};
1193#define NCPUMODES (sizeof(cpumodes)/sizeof(u8))
1194
1195static void ip__resolve_ams(struct machine *machine, struct thread *thread, 1192static void ip__resolve_ams(struct machine *machine, struct thread *thread,
1196 struct addr_map_symbol *ams, 1193 struct addr_map_symbol *ams,
1197 u64 ip) 1194 u64 ip)
1198{ 1195{
1199 struct addr_location al; 1196 struct addr_location al;
1200 size_t i;
1201 u8 m;
1202 1197
1203 memset(&al, 0, sizeof(al)); 1198 memset(&al, 0, sizeof(al));
1199 /*
1200 * We cannot use the header.misc hint to determine whether a
1201 * branch stack address is user, kernel, guest, hypervisor.
1202 * Branches may straddle the kernel/user/hypervisor boundaries.
1203 * Thus, we have to try consecutively until we find a match
1204 * or else, the symbol is unknown
1205 */
1206 thread__find_cpumode_addr_location(thread, machine, MAP__FUNCTION, ip, &al);
1204 1207
1205 for (i = 0; i < NCPUMODES; i++) {
1206 m = cpumodes[i];
1207 /*
1208 * We cannot use the header.misc hint to determine whether a
1209 * branch stack address is user, kernel, guest, hypervisor.
1210 * Branches may straddle the kernel/user/hypervisor boundaries.
1211 * Thus, we have to try consecutively until we find a match
1212 * or else, the symbol is unknown
1213 */
1214 thread__find_addr_location(thread, machine, m, MAP__FUNCTION,
1215 ip, &al);
1216 if (al.map)
1217 goto found;
1218 }
1219found:
1220 ams->addr = ip; 1208 ams->addr = ip;
1221 ams->al_addr = al.addr; 1209 ams->al_addr = al.addr;
1222 ams->sym = al.sym; 1210 ams->sym = al.sym;
@@ -1238,37 +1226,35 @@ static void ip__resolve_data(struct machine *machine, struct thread *thread,
1238 ams->map = al.map; 1226 ams->map = al.map;
1239} 1227}
1240 1228
1241struct mem_info *machine__resolve_mem(struct machine *machine, 1229struct mem_info *sample__resolve_mem(struct perf_sample *sample,
1242 struct thread *thr, 1230 struct addr_location *al)
1243 struct perf_sample *sample,
1244 u8 cpumode)
1245{ 1231{
1246 struct mem_info *mi = zalloc(sizeof(*mi)); 1232 struct mem_info *mi = zalloc(sizeof(*mi));
1247 1233
1248 if (!mi) 1234 if (!mi)
1249 return NULL; 1235 return NULL;
1250 1236
1251 ip__resolve_ams(machine, thr, &mi->iaddr, sample->ip); 1237 ip__resolve_ams(al->machine, al->thread, &mi->iaddr, sample->ip);
1252 ip__resolve_data(machine, thr, cpumode, &mi->daddr, sample->addr); 1238 ip__resolve_data(al->machine, al->thread, al->cpumode,
1239 &mi->daddr, sample->addr);
1253 mi->data_src.val = sample->data_src; 1240 mi->data_src.val = sample->data_src;
1254 1241
1255 return mi; 1242 return mi;
1256} 1243}
1257 1244
1258struct branch_info *machine__resolve_bstack(struct machine *machine, 1245struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
1259 struct thread *thr, 1246 struct addr_location *al)
1260 struct branch_stack *bs)
1261{ 1247{
1262 struct branch_info *bi;
1263 unsigned int i; 1248 unsigned int i;
1249 const struct branch_stack *bs = sample->branch_stack;
1250 struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
1264 1251
1265 bi = calloc(bs->nr, sizeof(struct branch_info));
1266 if (!bi) 1252 if (!bi)
1267 return NULL; 1253 return NULL;
1268 1254
1269 for (i = 0; i < bs->nr; i++) { 1255 for (i = 0; i < bs->nr; i++) {
1270 ip__resolve_ams(machine, thr, &bi[i].to, bs->entries[i].to); 1256 ip__resolve_ams(al->machine, al->thread, &bi[i].to, bs->entries[i].to);
1271 ip__resolve_ams(machine, thr, &bi[i].from, bs->entries[i].from); 1257 ip__resolve_ams(al->machine, al->thread, &bi[i].from, bs->entries[i].from);
1272 bi[i].flags = bs->entries[i].flags; 1258 bi[i].flags = bs->entries[i].flags;
1273 } 1259 }
1274 return bi; 1260 return bi;
@@ -1326,7 +1312,7 @@ static int machine__resolve_callchain_sample(struct machine *machine,
1326 continue; 1312 continue;
1327 } 1313 }
1328 1314
1329 al.filtered = false; 1315 al.filtered = 0;
1330 thread__find_addr_location(thread, machine, cpumode, 1316 thread__find_addr_location(thread, machine, cpumode,
1331 MAP__FUNCTION, ip, &al); 1317 MAP__FUNCTION, ip, &al);
1332 if (al.sym != NULL) { 1318 if (al.sym != NULL) {
@@ -1385,8 +1371,7 @@ int machine__resolve_callchain(struct machine *machine,
1385 return 0; 1371 return 0;
1386 1372
1387 return unwind__get_entries(unwind_entry, &callchain_cursor, machine, 1373 return unwind__get_entries(unwind_entry, &callchain_cursor, machine,
1388 thread, evsel->attr.sample_regs_user, 1374 thread, sample, max_stack);
1389 sample, max_stack);
1390 1375
1391} 1376}
1392 1377
diff --git a/tools/perf/util/machine.h b/tools/perf/util/machine.h
index f77e91e483dc..c8c74a119398 100644
--- a/tools/perf/util/machine.h
+++ b/tools/perf/util/machine.h
@@ -41,7 +41,8 @@ struct map *machine__kernel_map(struct machine *machine, enum map_type type)
41 return machine->vmlinux_maps[type]; 41 return machine->vmlinux_maps[type];
42} 42}
43 43
44struct thread *machine__find_thread(struct machine *machine, pid_t tid); 44struct thread *machine__find_thread(struct machine *machine, pid_t pid,
45 pid_t tid);
45 46
46int machine__process_comm_event(struct machine *machine, union perf_event *event, 47int machine__process_comm_event(struct machine *machine, union perf_event *event,
47 struct perf_sample *sample); 48 struct perf_sample *sample);
@@ -91,12 +92,10 @@ void machine__delete_dead_threads(struct machine *machine);
91void machine__delete_threads(struct machine *machine); 92void machine__delete_threads(struct machine *machine);
92void machine__delete(struct machine *machine); 93void machine__delete(struct machine *machine);
93 94
94struct branch_info *machine__resolve_bstack(struct machine *machine, 95struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
95 struct thread *thread, 96 struct addr_location *al);
96 struct branch_stack *bs); 97struct mem_info *sample__resolve_mem(struct perf_sample *sample,
97struct mem_info *machine__resolve_mem(struct machine *machine, 98 struct addr_location *al);
98 struct thread *thread,
99 struct perf_sample *sample, u8 cpumode);
100int machine__resolve_callchain(struct machine *machine, 99int machine__resolve_callchain(struct machine *machine,
101 struct perf_evsel *evsel, 100 struct perf_evsel *evsel,
102 struct thread *thread, 101 struct thread *thread,
diff --git a/tools/perf/util/map.h b/tools/perf/util/map.h
index 257e513205ce..f00f058afb3b 100644
--- a/tools/perf/util/map.h
+++ b/tools/perf/util/map.h
@@ -90,6 +90,16 @@ u64 map__objdump_2mem(struct map *map, u64 ip);
90 90
91struct symbol; 91struct symbol;
92 92
93/* map__for_each_symbol - iterate over the symbols in the given map
94 *
95 * @map: the 'struct map *' in which symbols itereated
96 * @pos: the 'struct symbol *' to use as a loop cursor
97 * @n: the 'struct rb_node *' to use as a temporary storage
98 * Note: caller must ensure map->dso is not NULL (map is loaded).
99 */
100#define map__for_each_symbol(map, pos, n) \
101 dso__for_each_symbol(map->dso, pos, n, map->type)
102
93typedef int (*symbol_filter_t)(struct map *map, struct symbol *sym); 103typedef int (*symbol_filter_t)(struct map *map, struct symbol *sym);
94 104
95void map__init(struct map *map, enum map_type type, 105void map__init(struct map *map, enum map_type type,
diff --git a/tools/perf/util/parse-options.c b/tools/perf/util/parse-options.c
index d22e3f8017dc..bf48092983c6 100644
--- a/tools/perf/util/parse-options.c
+++ b/tools/perf/util/parse-options.c
@@ -407,7 +407,9 @@ int parse_options_step(struct parse_opt_ctx_t *ctx,
407 if (internal_help && !strcmp(arg + 2, "help")) 407 if (internal_help && !strcmp(arg + 2, "help"))
408 return usage_with_options_internal(usagestr, options, 0); 408 return usage_with_options_internal(usagestr, options, 0);
409 if (!strcmp(arg + 2, "list-opts")) 409 if (!strcmp(arg + 2, "list-opts"))
410 return PARSE_OPT_LIST; 410 return PARSE_OPT_LIST_OPTS;
411 if (!strcmp(arg + 2, "list-cmds"))
412 return PARSE_OPT_LIST_SUBCMDS;
411 switch (parse_long_opt(ctx, arg + 2, options)) { 413 switch (parse_long_opt(ctx, arg + 2, options)) {
412 case -1: 414 case -1:
413 return parse_options_usage(usagestr, options, arg + 2, 0); 415 return parse_options_usage(usagestr, options, arg + 2, 0);
@@ -433,25 +435,45 @@ int parse_options_end(struct parse_opt_ctx_t *ctx)
433 return ctx->cpidx + ctx->argc; 435 return ctx->cpidx + ctx->argc;
434} 436}
435 437
436int parse_options(int argc, const char **argv, const struct option *options, 438int parse_options_subcommand(int argc, const char **argv, const struct option *options,
437 const char * const usagestr[], int flags) 439 const char *const subcommands[], const char *usagestr[], int flags)
438{ 440{
439 struct parse_opt_ctx_t ctx; 441 struct parse_opt_ctx_t ctx;
440 442
441 perf_header__set_cmdline(argc, argv); 443 perf_header__set_cmdline(argc, argv);
442 444
445 /* build usage string if it's not provided */
446 if (subcommands && !usagestr[0]) {
447 struct strbuf buf = STRBUF_INIT;
448
449 strbuf_addf(&buf, "perf %s [<options>] {", argv[0]);
450 for (int i = 0; subcommands[i]; i++) {
451 if (i)
452 strbuf_addstr(&buf, "|");
453 strbuf_addstr(&buf, subcommands[i]);
454 }
455 strbuf_addstr(&buf, "}");
456
457 usagestr[0] = strdup(buf.buf);
458 strbuf_release(&buf);
459 }
460
443 parse_options_start(&ctx, argc, argv, flags); 461 parse_options_start(&ctx, argc, argv, flags);
444 switch (parse_options_step(&ctx, options, usagestr)) { 462 switch (parse_options_step(&ctx, options, usagestr)) {
445 case PARSE_OPT_HELP: 463 case PARSE_OPT_HELP:
446 exit(129); 464 exit(129);
447 case PARSE_OPT_DONE: 465 case PARSE_OPT_DONE:
448 break; 466 break;
449 case PARSE_OPT_LIST: 467 case PARSE_OPT_LIST_OPTS:
450 while (options->type != OPTION_END) { 468 while (options->type != OPTION_END) {
451 printf("--%s ", options->long_name); 469 printf("--%s ", options->long_name);
452 options++; 470 options++;
453 } 471 }
454 exit(130); 472 exit(130);
473 case PARSE_OPT_LIST_SUBCMDS:
474 for (int i = 0; subcommands[i]; i++)
475 printf("%s ", subcommands[i]);
476 exit(130);
455 default: /* PARSE_OPT_UNKNOWN */ 477 default: /* PARSE_OPT_UNKNOWN */
456 if (ctx.argv[0][1] == '-') { 478 if (ctx.argv[0][1] == '-') {
457 error("unknown option `%s'", ctx.argv[0] + 2); 479 error("unknown option `%s'", ctx.argv[0] + 2);
@@ -464,6 +486,13 @@ int parse_options(int argc, const char **argv, const struct option *options,
464 return parse_options_end(&ctx); 486 return parse_options_end(&ctx);
465} 487}
466 488
489int parse_options(int argc, const char **argv, const struct option *options,
490 const char * const usagestr[], int flags)
491{
492 return parse_options_subcommand(argc, argv, options, NULL,
493 (const char **) usagestr, flags);
494}
495
467#define USAGE_OPTS_WIDTH 24 496#define USAGE_OPTS_WIDTH 24
468#define USAGE_GAP 2 497#define USAGE_GAP 2
469 498
diff --git a/tools/perf/util/parse-options.h b/tools/perf/util/parse-options.h
index cbf0149cf221..d8dac8ac5f37 100644
--- a/tools/perf/util/parse-options.h
+++ b/tools/perf/util/parse-options.h
@@ -140,6 +140,11 @@ extern int parse_options(int argc, const char **argv,
140 const struct option *options, 140 const struct option *options,
141 const char * const usagestr[], int flags); 141 const char * const usagestr[], int flags);
142 142
143extern int parse_options_subcommand(int argc, const char **argv,
144 const struct option *options,
145 const char *const subcommands[],
146 const char *usagestr[], int flags);
147
143extern NORETURN void usage_with_options(const char * const *usagestr, 148extern NORETURN void usage_with_options(const char * const *usagestr,
144 const struct option *options); 149 const struct option *options);
145 150
@@ -148,7 +153,8 @@ extern NORETURN void usage_with_options(const char * const *usagestr,
148enum { 153enum {
149 PARSE_OPT_HELP = -1, 154 PARSE_OPT_HELP = -1,
150 PARSE_OPT_DONE, 155 PARSE_OPT_DONE,
151 PARSE_OPT_LIST, 156 PARSE_OPT_LIST_OPTS,
157 PARSE_OPT_LIST_SUBCMDS,
152 PARSE_OPT_UNKNOWN, 158 PARSE_OPT_UNKNOWN,
153}; 159};
154 160
diff --git a/tools/perf/util/perf_regs.c b/tools/perf/util/perf_regs.c
new file mode 100644
index 000000000000..a3539ef30b15
--- /dev/null
+++ b/tools/perf/util/perf_regs.c
@@ -0,0 +1,19 @@
1#include <errno.h>
2#include "perf_regs.h"
3
4int perf_reg_value(u64 *valp, struct regs_dump *regs, int id)
5{
6 int i, idx = 0;
7 u64 mask = regs->mask;
8
9 if (!(mask & (1 << id)))
10 return -EINVAL;
11
12 for (i = 0; i < id; i++) {
13 if (mask & (1 << i))
14 idx++;
15 }
16
17 *valp = regs->regs[idx];
18 return 0;
19}
diff --git a/tools/perf/util/perf_regs.h b/tools/perf/util/perf_regs.h
index a3d42cd74919..d6e8b6a8d7f3 100644
--- a/tools/perf/util/perf_regs.h
+++ b/tools/perf/util/perf_regs.h
@@ -1,8 +1,14 @@
1#ifndef __PERF_REGS_H 1#ifndef __PERF_REGS_H
2#define __PERF_REGS_H 2#define __PERF_REGS_H
3 3
4#include "types.h"
5#include "event.h"
6
4#ifdef HAVE_PERF_REGS_SUPPORT 7#ifdef HAVE_PERF_REGS_SUPPORT
5#include <perf_regs.h> 8#include <perf_regs.h>
9
10int perf_reg_value(u64 *valp, struct regs_dump *regs, int id);
11
6#else 12#else
7#define PERF_REGS_MASK 0 13#define PERF_REGS_MASK 0
8 14
@@ -10,5 +16,12 @@ static inline const char *perf_reg_name(int id __maybe_unused)
10{ 16{
11 return NULL; 17 return NULL;
12} 18}
19
20static inline int perf_reg_value(u64 *valp __maybe_unused,
21 struct regs_dump *regs __maybe_unused,
22 int id __maybe_unused)
23{
24 return 0;
25}
13#endif /* HAVE_PERF_REGS_SUPPORT */ 26#endif /* HAVE_PERF_REGS_SUPPORT */
14#endif /* __PERF_REGS_H */ 27#endif /* __PERF_REGS_H */
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
index b752ecb40d86..00a7dcb2f55c 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -3,7 +3,7 @@
3#include <unistd.h> 3#include <unistd.h>
4#include <stdio.h> 4#include <stdio.h>
5#include <dirent.h> 5#include <dirent.h>
6#include "fs.h" 6#include <api/fs/fs.h>
7#include <locale.h> 7#include <locale.h>
8#include "util.h" 8#include "util.h"
9#include "pmu.h" 9#include "pmu.h"
diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
index d8b048c20cde..0d1542f33d87 100644
--- a/tools/perf/util/probe-event.c
+++ b/tools/perf/util/probe-event.c
@@ -70,34 +70,32 @@ static int e_snprintf(char *str, size_t size, const char *format, ...)
70} 70}
71 71
72static char *synthesize_perf_probe_point(struct perf_probe_point *pp); 72static char *synthesize_perf_probe_point(struct perf_probe_point *pp);
73static int convert_name_to_addr(struct perf_probe_event *pev,
74 const char *exec);
75static void clear_probe_trace_event(struct probe_trace_event *tev); 73static void clear_probe_trace_event(struct probe_trace_event *tev);
76static struct machine machine; 74static struct machine *host_machine;
77 75
78/* Initialize symbol maps and path of vmlinux/modules */ 76/* Initialize symbol maps and path of vmlinux/modules */
79static int init_vmlinux(void) 77static int init_symbol_maps(bool user_only)
80{ 78{
81 int ret; 79 int ret;
82 80
83 symbol_conf.sort_by_name = true; 81 symbol_conf.sort_by_name = true;
84 if (symbol_conf.vmlinux_name == NULL)
85 symbol_conf.try_vmlinux_path = true;
86 else
87 pr_debug("Use vmlinux: %s\n", symbol_conf.vmlinux_name);
88 ret = symbol__init(); 82 ret = symbol__init();
89 if (ret < 0) { 83 if (ret < 0) {
90 pr_debug("Failed to init symbol map.\n"); 84 pr_debug("Failed to init symbol map.\n");
91 goto out; 85 goto out;
92 } 86 }
93 87
94 ret = machine__init(&machine, "", HOST_KERNEL_ID); 88 if (host_machine || user_only) /* already initialized */
95 if (ret < 0) 89 return 0;
96 goto out;
97 90
98 if (machine__create_kernel_maps(&machine) < 0) { 91 if (symbol_conf.vmlinux_name)
99 pr_debug("machine__create_kernel_maps() failed.\n"); 92 pr_debug("Use vmlinux: %s\n", symbol_conf.vmlinux_name);
100 goto out; 93
94 host_machine = machine__new_host();
95 if (!host_machine) {
96 pr_debug("machine__new_host() failed.\n");
97 symbol__exit();
98 ret = -1;
101 } 99 }
102out: 100out:
103 if (ret < 0) 101 if (ret < 0)
@@ -105,21 +103,66 @@ out:
105 return ret; 103 return ret;
106} 104}
107 105
106static void exit_symbol_maps(void)
107{
108 if (host_machine) {
109 machine__delete(host_machine);
110 host_machine = NULL;
111 }
112 symbol__exit();
113}
114
108static struct symbol *__find_kernel_function_by_name(const char *name, 115static struct symbol *__find_kernel_function_by_name(const char *name,
109 struct map **mapp) 116 struct map **mapp)
110{ 117{
111 return machine__find_kernel_function_by_name(&machine, name, mapp, 118 return machine__find_kernel_function_by_name(host_machine, name, mapp,
112 NULL); 119 NULL);
113} 120}
114 121
122static struct symbol *__find_kernel_function(u64 addr, struct map **mapp)
123{
124 return machine__find_kernel_function(host_machine, addr, mapp, NULL);
125}
126
127static struct ref_reloc_sym *kernel_get_ref_reloc_sym(void)
128{
129 /* kmap->ref_reloc_sym should be set if host_machine is initialized */
130 struct kmap *kmap;
131
132 if (map__load(host_machine->vmlinux_maps[MAP__FUNCTION], NULL) < 0)
133 return NULL;
134
135 kmap = map__kmap(host_machine->vmlinux_maps[MAP__FUNCTION]);
136 return kmap->ref_reloc_sym;
137}
138
139static u64 kernel_get_symbol_address_by_name(const char *name, bool reloc)
140{
141 struct ref_reloc_sym *reloc_sym;
142 struct symbol *sym;
143 struct map *map;
144
145 /* ref_reloc_sym is just a label. Need a special fix*/
146 reloc_sym = kernel_get_ref_reloc_sym();
147 if (reloc_sym && strcmp(name, reloc_sym->name) == 0)
148 return (reloc) ? reloc_sym->addr : reloc_sym->unrelocated_addr;
149 else {
150 sym = __find_kernel_function_by_name(name, &map);
151 if (sym)
152 return map->unmap_ip(map, sym->start) -
153 (reloc) ? 0 : map->reloc;
154 }
155 return 0;
156}
157
115static struct map *kernel_get_module_map(const char *module) 158static struct map *kernel_get_module_map(const char *module)
116{ 159{
117 struct rb_node *nd; 160 struct rb_node *nd;
118 struct map_groups *grp = &machine.kmaps; 161 struct map_groups *grp = &host_machine->kmaps;
119 162
120 /* A file path -- this is an offline module */ 163 /* A file path -- this is an offline module */
121 if (module && strchr(module, '/')) 164 if (module && strchr(module, '/'))
122 return machine__new_module(&machine, 0, module); 165 return machine__new_module(host_machine, 0, module);
123 166
124 if (!module) 167 if (!module)
125 module = "kernel"; 168 module = "kernel";
@@ -141,7 +184,7 @@ static struct dso *kernel_get_module_dso(const char *module)
141 const char *vmlinux_name; 184 const char *vmlinux_name;
142 185
143 if (module) { 186 if (module) {
144 list_for_each_entry(dso, &machine.kernel_dsos, node) { 187 list_for_each_entry(dso, &host_machine->kernel_dsos, node) {
145 if (strncmp(dso->short_name + 1, module, 188 if (strncmp(dso->short_name + 1, module,
146 dso->short_name_len - 2) == 0) 189 dso->short_name_len - 2) == 0)
147 goto found; 190 goto found;
@@ -150,7 +193,7 @@ static struct dso *kernel_get_module_dso(const char *module)
150 return NULL; 193 return NULL;
151 } 194 }
152 195
153 map = machine.vmlinux_maps[MAP__FUNCTION]; 196 map = host_machine->vmlinux_maps[MAP__FUNCTION];
154 dso = map->dso; 197 dso = map->dso;
155 198
156 vmlinux_name = symbol_conf.vmlinux_name; 199 vmlinux_name = symbol_conf.vmlinux_name;
@@ -173,20 +216,6 @@ const char *kernel_get_module_path(const char *module)
173 return (dso) ? dso->long_name : NULL; 216 return (dso) ? dso->long_name : NULL;
174} 217}
175 218
176static int init_user_exec(void)
177{
178 int ret = 0;
179
180 symbol_conf.try_vmlinux_path = false;
181 symbol_conf.sort_by_name = true;
182 ret = symbol__init();
183
184 if (ret < 0)
185 pr_debug("Failed to init symbol map.\n");
186
187 return ret;
188}
189
190static int convert_exec_to_group(const char *exec, char **result) 219static int convert_exec_to_group(const char *exec, char **result)
191{ 220{
192 char *ptr1, *ptr2, *exec_copy; 221 char *ptr1, *ptr2, *exec_copy;
@@ -218,32 +247,23 @@ out:
218 return ret; 247 return ret;
219} 248}
220 249
221static int convert_to_perf_probe_point(struct probe_trace_point *tp, 250static void clear_probe_trace_events(struct probe_trace_event *tevs, int ntevs)
222 struct perf_probe_point *pp)
223{ 251{
224 pp->function = strdup(tp->symbol); 252 int i;
225
226 if (pp->function == NULL)
227 return -ENOMEM;
228
229 pp->offset = tp->offset;
230 pp->retprobe = tp->retprobe;
231 253
232 return 0; 254 for (i = 0; i < ntevs; i++)
255 clear_probe_trace_event(tevs + i);
233} 256}
234 257
235#ifdef HAVE_DWARF_SUPPORT 258#ifdef HAVE_DWARF_SUPPORT
259
236/* Open new debuginfo of given module */ 260/* Open new debuginfo of given module */
237static struct debuginfo *open_debuginfo(const char *module) 261static struct debuginfo *open_debuginfo(const char *module)
238{ 262{
239 const char *path; 263 const char *path = module;
240 264
241 /* A file path -- this is an offline module */ 265 if (!module || !strchr(module, '/')) {
242 if (module && strchr(module, '/'))
243 path = module;
244 else {
245 path = kernel_get_module_path(module); 266 path = kernel_get_module_path(module);
246
247 if (!path) { 267 if (!path) {
248 pr_err("Failed to find path of %s module.\n", 268 pr_err("Failed to find path of %s module.\n",
249 module ?: "kernel"); 269 module ?: "kernel");
@@ -253,46 +273,6 @@ static struct debuginfo *open_debuginfo(const char *module)
253 return debuginfo__new(path); 273 return debuginfo__new(path);
254} 274}
255 275
256/*
257 * Convert trace point to probe point with debuginfo
258 * Currently only handles kprobes.
259 */
260static int kprobe_convert_to_perf_probe(struct probe_trace_point *tp,
261 struct perf_probe_point *pp)
262{
263 struct symbol *sym;
264 struct map *map;
265 u64 addr;
266 int ret = -ENOENT;
267 struct debuginfo *dinfo;
268
269 sym = __find_kernel_function_by_name(tp->symbol, &map);
270 if (sym) {
271 addr = map->unmap_ip(map, sym->start + tp->offset);
272 pr_debug("try to find %s+%ld@%" PRIx64 "\n", tp->symbol,
273 tp->offset, addr);
274
275 dinfo = debuginfo__new_online_kernel(addr);
276 if (dinfo) {
277 ret = debuginfo__find_probe_point(dinfo,
278 (unsigned long)addr, pp);
279 debuginfo__delete(dinfo);
280 } else {
281 pr_debug("Failed to open debuginfo at 0x%" PRIx64 "\n",
282 addr);
283 ret = -ENOENT;
284 }
285 }
286 if (ret <= 0) {
287 pr_debug("Failed to find corresponding probes from "
288 "debuginfo. Use kprobe event information.\n");
289 return convert_to_perf_probe_point(tp, pp);
290 }
291 pp->retprobe = tp->retprobe;
292
293 return 0;
294}
295
296static int get_text_start_address(const char *exec, unsigned long *address) 276static int get_text_start_address(const char *exec, unsigned long *address)
297{ 277{
298 Elf *elf; 278 Elf *elf;
@@ -321,12 +301,62 @@ out:
321 return ret; 301 return ret;
322} 302}
323 303
304/*
305 * Convert trace point to probe point with debuginfo
306 */
307static int find_perf_probe_point_from_dwarf(struct probe_trace_point *tp,
308 struct perf_probe_point *pp,
309 bool is_kprobe)
310{
311 struct debuginfo *dinfo = NULL;
312 unsigned long stext = 0;
313 u64 addr = tp->address;
314 int ret = -ENOENT;
315
316 /* convert the address to dwarf address */
317 if (!is_kprobe) {
318 if (!addr) {
319 ret = -EINVAL;
320 goto error;
321 }
322 ret = get_text_start_address(tp->module, &stext);
323 if (ret < 0)
324 goto error;
325 addr += stext;
326 } else {
327 addr = kernel_get_symbol_address_by_name(tp->symbol, false);
328 if (addr == 0)
329 goto error;
330 addr += tp->offset;
331 }
332
333 pr_debug("try to find information at %" PRIx64 " in %s\n", addr,
334 tp->module ? : "kernel");
335
336 dinfo = open_debuginfo(tp->module);
337 if (dinfo) {
338 ret = debuginfo__find_probe_point(dinfo,
339 (unsigned long)addr, pp);
340 debuginfo__delete(dinfo);
341 } else {
342 pr_debug("Failed to open debuginfo at 0x%" PRIx64 "\n", addr);
343 ret = -ENOENT;
344 }
345
346 if (ret > 0) {
347 pp->retprobe = tp->retprobe;
348 return 0;
349 }
350error:
351 pr_debug("Failed to find corresponding probes from debuginfo.\n");
352 return ret ? : -ENOENT;
353}
354
324static int add_exec_to_probe_trace_events(struct probe_trace_event *tevs, 355static int add_exec_to_probe_trace_events(struct probe_trace_event *tevs,
325 int ntevs, const char *exec) 356 int ntevs, const char *exec)
326{ 357{
327 int i, ret = 0; 358 int i, ret = 0;
328 unsigned long offset, stext = 0; 359 unsigned long stext = 0;
329 char buf[32];
330 360
331 if (!exec) 361 if (!exec)
332 return 0; 362 return 0;
@@ -337,15 +367,9 @@ static int add_exec_to_probe_trace_events(struct probe_trace_event *tevs,
337 367
338 for (i = 0; i < ntevs && ret >= 0; i++) { 368 for (i = 0; i < ntevs && ret >= 0; i++) {
339 /* point.address is the addres of point.symbol + point.offset */ 369 /* point.address is the addres of point.symbol + point.offset */
340 offset = tevs[i].point.address - stext; 370 tevs[i].point.address -= stext;
341 tevs[i].point.offset = 0;
342 zfree(&tevs[i].point.symbol);
343 ret = e_snprintf(buf, 32, "0x%lx", offset);
344 if (ret < 0)
345 break;
346 tevs[i].point.module = strdup(exec); 371 tevs[i].point.module = strdup(exec);
347 tevs[i].point.symbol = strdup(buf); 372 if (!tevs[i].point.module) {
348 if (!tevs[i].point.symbol || !tevs[i].point.module) {
349 ret = -ENOMEM; 373 ret = -ENOMEM;
350 break; 374 break;
351 } 375 }
@@ -388,12 +412,40 @@ static int add_module_to_probe_trace_events(struct probe_trace_event *tevs,
388 return ret; 412 return ret;
389} 413}
390 414
391static void clear_probe_trace_events(struct probe_trace_event *tevs, int ntevs) 415/* Post processing the probe events */
416static int post_process_probe_trace_events(struct probe_trace_event *tevs,
417 int ntevs, const char *module,
418 bool uprobe)
392{ 419{
420 struct ref_reloc_sym *reloc_sym;
421 char *tmp;
393 int i; 422 int i;
394 423
395 for (i = 0; i < ntevs; i++) 424 if (uprobe)
396 clear_probe_trace_event(tevs + i); 425 return add_exec_to_probe_trace_events(tevs, ntevs, module);
426
427 /* Note that currently ref_reloc_sym based probe is not for drivers */
428 if (module)
429 return add_module_to_probe_trace_events(tevs, ntevs, module);
430
431 reloc_sym = kernel_get_ref_reloc_sym();
432 if (!reloc_sym) {
433 pr_warning("Relocated base symbol is not found!\n");
434 return -EINVAL;
435 }
436
437 for (i = 0; i < ntevs; i++) {
438 if (tevs[i].point.address) {
439 tmp = strdup(reloc_sym->name);
440 if (!tmp)
441 return -ENOMEM;
442 free(tevs[i].point.symbol);
443 tevs[i].point.symbol = tmp;
444 tevs[i].point.offset = tevs[i].point.address -
445 reloc_sym->unrelocated_addr;
446 }
447 }
448 return 0;
397} 449}
398 450
399/* Try to find perf_probe_event with debuginfo */ 451/* Try to find perf_probe_event with debuginfo */
@@ -416,21 +468,16 @@ static int try_to_find_probe_trace_events(struct perf_probe_event *pev,
416 return 0; 468 return 0;
417 } 469 }
418 470
471 pr_debug("Try to find probe point from debuginfo.\n");
419 /* Searching trace events corresponding to a probe event */ 472 /* Searching trace events corresponding to a probe event */
420 ntevs = debuginfo__find_trace_events(dinfo, pev, tevs, max_tevs); 473 ntevs = debuginfo__find_trace_events(dinfo, pev, tevs, max_tevs);
421 474
422 debuginfo__delete(dinfo); 475 debuginfo__delete(dinfo);
423 476
424 if (ntevs > 0) { /* Succeeded to find trace events */ 477 if (ntevs > 0) { /* Succeeded to find trace events */
425 pr_debug("find %d probe_trace_events.\n", ntevs); 478 pr_debug("Found %d probe_trace_events.\n", ntevs);
426 if (target) { 479 ret = post_process_probe_trace_events(*tevs, ntevs,
427 if (pev->uprobes) 480 target, pev->uprobes);
428 ret = add_exec_to_probe_trace_events(*tevs,
429 ntevs, target);
430 else
431 ret = add_module_to_probe_trace_events(*tevs,
432 ntevs, target);
433 }
434 if (ret < 0) { 481 if (ret < 0) {
435 clear_probe_trace_events(*tevs, ntevs); 482 clear_probe_trace_events(*tevs, ntevs);
436 zfree(tevs); 483 zfree(tevs);
@@ -563,20 +610,16 @@ static int _show_one_line(FILE *fp, int l, bool skip, bool show_num)
563 * Show line-range always requires debuginfo to find source file and 610 * Show line-range always requires debuginfo to find source file and
564 * line number. 611 * line number.
565 */ 612 */
566int show_line_range(struct line_range *lr, const char *module) 613static int __show_line_range(struct line_range *lr, const char *module)
567{ 614{
568 int l = 1; 615 int l = 1;
569 struct line_node *ln; 616 struct int_node *ln;
570 struct debuginfo *dinfo; 617 struct debuginfo *dinfo;
571 FILE *fp; 618 FILE *fp;
572 int ret; 619 int ret;
573 char *tmp; 620 char *tmp;
574 621
575 /* Search a line range */ 622 /* Search a line range */
576 ret = init_vmlinux();
577 if (ret < 0)
578 return ret;
579
580 dinfo = open_debuginfo(module); 623 dinfo = open_debuginfo(module);
581 if (!dinfo) { 624 if (!dinfo) {
582 pr_warning("Failed to open debuginfo file.\n"); 625 pr_warning("Failed to open debuginfo file.\n");
@@ -623,8 +666,8 @@ int show_line_range(struct line_range *lr, const char *module)
623 goto end; 666 goto end;
624 } 667 }
625 668
626 list_for_each_entry(ln, &lr->line_list, list) { 669 intlist__for_each(ln, lr->line_list) {
627 for (; ln->line > l; l++) { 670 for (; ln->i > l; l++) {
628 ret = show_one_line(fp, l - lr->offset); 671 ret = show_one_line(fp, l - lr->offset);
629 if (ret < 0) 672 if (ret < 0)
630 goto end; 673 goto end;
@@ -646,6 +689,19 @@ end:
646 return ret; 689 return ret;
647} 690}
648 691
692int show_line_range(struct line_range *lr, const char *module)
693{
694 int ret;
695
696 ret = init_symbol_maps(false);
697 if (ret < 0)
698 return ret;
699 ret = __show_line_range(lr, module);
700 exit_symbol_maps();
701
702 return ret;
703}
704
649static int show_available_vars_at(struct debuginfo *dinfo, 705static int show_available_vars_at(struct debuginfo *dinfo,
650 struct perf_probe_event *pev, 706 struct perf_probe_event *pev,
651 int max_vls, struct strfilter *_filter, 707 int max_vls, struct strfilter *_filter,
@@ -707,14 +763,15 @@ int show_available_vars(struct perf_probe_event *pevs, int npevs,
707 int i, ret = 0; 763 int i, ret = 0;
708 struct debuginfo *dinfo; 764 struct debuginfo *dinfo;
709 765
710 ret = init_vmlinux(); 766 ret = init_symbol_maps(false);
711 if (ret < 0) 767 if (ret < 0)
712 return ret; 768 return ret;
713 769
714 dinfo = open_debuginfo(module); 770 dinfo = open_debuginfo(module);
715 if (!dinfo) { 771 if (!dinfo) {
716 pr_warning("Failed to open debuginfo file.\n"); 772 pr_warning("Failed to open debuginfo file.\n");
717 return -ENOENT; 773 ret = -ENOENT;
774 goto out;
718 } 775 }
719 776
720 setup_pager(); 777 setup_pager();
@@ -724,23 +781,19 @@ int show_available_vars(struct perf_probe_event *pevs, int npevs,
724 externs); 781 externs);
725 782
726 debuginfo__delete(dinfo); 783 debuginfo__delete(dinfo);
784out:
785 exit_symbol_maps();
727 return ret; 786 return ret;
728} 787}
729 788
730#else /* !HAVE_DWARF_SUPPORT */ 789#else /* !HAVE_DWARF_SUPPORT */
731 790
732static int kprobe_convert_to_perf_probe(struct probe_trace_point *tp, 791static int
733 struct perf_probe_point *pp) 792find_perf_probe_point_from_dwarf(struct probe_trace_point *tp __maybe_unused,
793 struct perf_probe_point *pp __maybe_unused,
794 bool is_kprobe __maybe_unused)
734{ 795{
735 struct symbol *sym; 796 return -ENOSYS;
736
737 sym = __find_kernel_function_by_name(tp->symbol, NULL);
738 if (!sym) {
739 pr_err("Failed to find symbol %s in kernel.\n", tp->symbol);
740 return -ENOENT;
741 }
742
743 return convert_to_perf_probe_point(tp, pp);
744} 797}
745 798
746static int try_to_find_probe_trace_events(struct perf_probe_event *pev, 799static int try_to_find_probe_trace_events(struct perf_probe_event *pev,
@@ -776,24 +829,22 @@ int show_available_vars(struct perf_probe_event *pevs __maybe_unused,
776 829
777void line_range__clear(struct line_range *lr) 830void line_range__clear(struct line_range *lr)
778{ 831{
779 struct line_node *ln;
780
781 free(lr->function); 832 free(lr->function);
782 free(lr->file); 833 free(lr->file);
783 free(lr->path); 834 free(lr->path);
784 free(lr->comp_dir); 835 free(lr->comp_dir);
785 while (!list_empty(&lr->line_list)) { 836 intlist__delete(lr->line_list);
786 ln = list_first_entry(&lr->line_list, struct line_node, list);
787 list_del(&ln->list);
788 free(ln);
789 }
790 memset(lr, 0, sizeof(*lr)); 837 memset(lr, 0, sizeof(*lr));
791} 838}
792 839
793void line_range__init(struct line_range *lr) 840int line_range__init(struct line_range *lr)
794{ 841{
795 memset(lr, 0, sizeof(*lr)); 842 memset(lr, 0, sizeof(*lr));
796 INIT_LIST_HEAD(&lr->line_list); 843 lr->line_list = intlist__new(NULL);
844 if (!lr->line_list)
845 return -ENOMEM;
846 else
847 return 0;
797} 848}
798 849
799static int parse_line_num(char **ptr, int *val, const char *what) 850static int parse_line_num(char **ptr, int *val, const char *what)
@@ -1267,16 +1318,21 @@ static int parse_probe_trace_command(const char *cmd,
1267 } else 1318 } else
1268 p = argv[1]; 1319 p = argv[1];
1269 fmt1_str = strtok_r(p, "+", &fmt); 1320 fmt1_str = strtok_r(p, "+", &fmt);
1270 tp->symbol = strdup(fmt1_str); 1321 if (fmt1_str[0] == '0') /* only the address started with 0x */
1271 if (tp->symbol == NULL) { 1322 tp->address = strtoul(fmt1_str, NULL, 0);
1272 ret = -ENOMEM; 1323 else {
1273 goto out; 1324 /* Only the symbol-based probe has offset */
1325 tp->symbol = strdup(fmt1_str);
1326 if (tp->symbol == NULL) {
1327 ret = -ENOMEM;
1328 goto out;
1329 }
1330 fmt2_str = strtok_r(NULL, "", &fmt);
1331 if (fmt2_str == NULL)
1332 tp->offset = 0;
1333 else
1334 tp->offset = strtoul(fmt2_str, NULL, 10);
1274 } 1335 }
1275 fmt2_str = strtok_r(NULL, "", &fmt);
1276 if (fmt2_str == NULL)
1277 tp->offset = 0;
1278 else
1279 tp->offset = strtoul(fmt2_str, NULL, 10);
1280 1336
1281 tev->nargs = argc - 2; 1337 tev->nargs = argc - 2;
1282 tev->args = zalloc(sizeof(struct probe_trace_arg) * tev->nargs); 1338 tev->args = zalloc(sizeof(struct probe_trace_arg) * tev->nargs);
@@ -1518,20 +1574,27 @@ char *synthesize_probe_trace_command(struct probe_trace_event *tev)
1518 if (buf == NULL) 1574 if (buf == NULL)
1519 return NULL; 1575 return NULL;
1520 1576
1577 len = e_snprintf(buf, MAX_CMDLEN, "%c:%s/%s ", tp->retprobe ? 'r' : 'p',
1578 tev->group, tev->event);
1579 if (len <= 0)
1580 goto error;
1581
1582 /* Uprobes must have tp->address and tp->module */
1583 if (tev->uprobes && (!tp->address || !tp->module))
1584 goto error;
1585
1586 /* Use the tp->address for uprobes */
1521 if (tev->uprobes) 1587 if (tev->uprobes)
1522 len = e_snprintf(buf, MAX_CMDLEN, "%c:%s/%s %s:%s", 1588 ret = e_snprintf(buf + len, MAX_CMDLEN - len, "%s:0x%lx",
1523 tp->retprobe ? 'r' : 'p', 1589 tp->module, tp->address);
1524 tev->group, tev->event,
1525 tp->module, tp->symbol);
1526 else 1590 else
1527 len = e_snprintf(buf, MAX_CMDLEN, "%c:%s/%s %s%s%s+%lu", 1591 ret = e_snprintf(buf + len, MAX_CMDLEN - len, "%s%s%s+%lu",
1528 tp->retprobe ? 'r' : 'p',
1529 tev->group, tev->event,
1530 tp->module ?: "", tp->module ? ":" : "", 1592 tp->module ?: "", tp->module ? ":" : "",
1531 tp->symbol, tp->offset); 1593 tp->symbol, tp->offset);
1532 1594
1533 if (len <= 0) 1595 if (ret <= 0)
1534 goto error; 1596 goto error;
1597 len += ret;
1535 1598
1536 for (i = 0; i < tev->nargs; i++) { 1599 for (i = 0; i < tev->nargs; i++) {
1537 ret = synthesize_probe_trace_arg(&tev->args[i], buf + len, 1600 ret = synthesize_probe_trace_arg(&tev->args[i], buf + len,
@@ -1547,6 +1610,79 @@ error:
1547 return NULL; 1610 return NULL;
1548} 1611}
1549 1612
1613static int find_perf_probe_point_from_map(struct probe_trace_point *tp,
1614 struct perf_probe_point *pp,
1615 bool is_kprobe)
1616{
1617 struct symbol *sym = NULL;
1618 struct map *map;
1619 u64 addr;
1620 int ret = -ENOENT;
1621
1622 if (!is_kprobe) {
1623 map = dso__new_map(tp->module);
1624 if (!map)
1625 goto out;
1626 addr = tp->address;
1627 sym = map__find_symbol(map, addr, NULL);
1628 } else {
1629 addr = kernel_get_symbol_address_by_name(tp->symbol, true);
1630 if (addr) {
1631 addr += tp->offset;
1632 sym = __find_kernel_function(addr, &map);
1633 }
1634 }
1635 if (!sym)
1636 goto out;
1637
1638 pp->retprobe = tp->retprobe;
1639 pp->offset = addr - map->unmap_ip(map, sym->start);
1640 pp->function = strdup(sym->name);
1641 ret = pp->function ? 0 : -ENOMEM;
1642
1643out:
1644 if (map && !is_kprobe) {
1645 dso__delete(map->dso);
1646 map__delete(map);
1647 }
1648
1649 return ret;
1650}
1651
1652static int convert_to_perf_probe_point(struct probe_trace_point *tp,
1653 struct perf_probe_point *pp,
1654 bool is_kprobe)
1655{
1656 char buf[128];
1657 int ret;
1658
1659 ret = find_perf_probe_point_from_dwarf(tp, pp, is_kprobe);
1660 if (!ret)
1661 return 0;
1662 ret = find_perf_probe_point_from_map(tp, pp, is_kprobe);
1663 if (!ret)
1664 return 0;
1665
1666 pr_debug("Failed to find probe point from both of dwarf and map.\n");
1667
1668 if (tp->symbol) {
1669 pp->function = strdup(tp->symbol);
1670 pp->offset = tp->offset;
1671 } else if (!tp->module && !is_kprobe) {
1672 ret = e_snprintf(buf, 128, "0x%" PRIx64, (u64)tp->address);
1673 if (ret < 0)
1674 return ret;
1675 pp->function = strdup(buf);
1676 pp->offset = 0;
1677 }
1678 if (pp->function == NULL)
1679 return -ENOMEM;
1680
1681 pp->retprobe = tp->retprobe;
1682
1683 return 0;
1684}
1685
1550static int convert_to_perf_probe_event(struct probe_trace_event *tev, 1686static int convert_to_perf_probe_event(struct probe_trace_event *tev,
1551 struct perf_probe_event *pev, bool is_kprobe) 1687 struct perf_probe_event *pev, bool is_kprobe)
1552{ 1688{
@@ -1560,11 +1696,7 @@ static int convert_to_perf_probe_event(struct probe_trace_event *tev,
1560 return -ENOMEM; 1696 return -ENOMEM;
1561 1697
1562 /* Convert trace_point to probe_point */ 1698 /* Convert trace_point to probe_point */
1563 if (is_kprobe) 1699 ret = convert_to_perf_probe_point(&tev->point, &pev->point, is_kprobe);
1564 ret = kprobe_convert_to_perf_probe(&tev->point, &pev->point);
1565 else
1566 ret = convert_to_perf_probe_point(&tev->point, &pev->point);
1567
1568 if (ret < 0) 1700 if (ret < 0)
1569 return ret; 1701 return ret;
1570 1702
@@ -1731,7 +1863,8 @@ static struct strlist *get_probe_trace_command_rawlist(int fd)
1731} 1863}
1732 1864
1733/* Show an event */ 1865/* Show an event */
1734static int show_perf_probe_event(struct perf_probe_event *pev) 1866static int show_perf_probe_event(struct perf_probe_event *pev,
1867 const char *module)
1735{ 1868{
1736 int i, ret; 1869 int i, ret;
1737 char buf[128]; 1870 char buf[128];
@@ -1747,6 +1880,8 @@ static int show_perf_probe_event(struct perf_probe_event *pev)
1747 return ret; 1880 return ret;
1748 1881
1749 printf(" %-20s (on %s", buf, place); 1882 printf(" %-20s (on %s", buf, place);
1883 if (module)
1884 printf(" in %s", module);
1750 1885
1751 if (pev->nargs > 0) { 1886 if (pev->nargs > 0) {
1752 printf(" with"); 1887 printf(" with");
@@ -1784,7 +1919,8 @@ static int __show_perf_probe_events(int fd, bool is_kprobe)
1784 ret = convert_to_perf_probe_event(&tev, &pev, 1919 ret = convert_to_perf_probe_event(&tev, &pev,
1785 is_kprobe); 1920 is_kprobe);
1786 if (ret >= 0) 1921 if (ret >= 0)
1787 ret = show_perf_probe_event(&pev); 1922 ret = show_perf_probe_event(&pev,
1923 tev.point.module);
1788 } 1924 }
1789 clear_perf_probe_event(&pev); 1925 clear_perf_probe_event(&pev);
1790 clear_probe_trace_event(&tev); 1926 clear_probe_trace_event(&tev);
@@ -1807,7 +1943,7 @@ int show_perf_probe_events(void)
1807 if (fd < 0) 1943 if (fd < 0)
1808 return fd; 1944 return fd;
1809 1945
1810 ret = init_vmlinux(); 1946 ret = init_symbol_maps(false);
1811 if (ret < 0) 1947 if (ret < 0)
1812 return ret; 1948 return ret;
1813 1949
@@ -1820,6 +1956,7 @@ int show_perf_probe_events(void)
1820 close(fd); 1956 close(fd);
1821 } 1957 }
1822 1958
1959 exit_symbol_maps();
1823 return ret; 1960 return ret;
1824} 1961}
1825 1962
@@ -1982,7 +2119,7 @@ static int __add_probe_trace_events(struct perf_probe_event *pev,
1982 group = pev->group; 2119 group = pev->group;
1983 pev->event = tev->event; 2120 pev->event = tev->event;
1984 pev->group = tev->group; 2121 pev->group = tev->group;
1985 show_perf_probe_event(pev); 2122 show_perf_probe_event(pev, tev->point.module);
1986 /* Trick here - restore current event/group */ 2123 /* Trick here - restore current event/group */
1987 pev->event = (char *)event; 2124 pev->event = (char *)event;
1988 pev->group = (char *)group; 2125 pev->group = (char *)group;
@@ -2008,113 +2145,175 @@ static int __add_probe_trace_events(struct perf_probe_event *pev,
2008 return ret; 2145 return ret;
2009} 2146}
2010 2147
2011static int convert_to_probe_trace_events(struct perf_probe_event *pev, 2148static char *looking_function_name;
2012 struct probe_trace_event **tevs, 2149static int num_matched_functions;
2013 int max_tevs, const char *target) 2150
2151static int probe_function_filter(struct map *map __maybe_unused,
2152 struct symbol *sym)
2014{ 2153{
2154 if ((sym->binding == STB_GLOBAL || sym->binding == STB_LOCAL) &&
2155 strcmp(looking_function_name, sym->name) == 0) {
2156 num_matched_functions++;
2157 return 0;
2158 }
2159 return 1;
2160}
2161
2162#define strdup_or_goto(str, label) \
2163 ({ char *__p = strdup(str); if (!__p) goto label; __p; })
2164
2165/*
2166 * Find probe function addresses from map.
2167 * Return an error or the number of found probe_trace_event
2168 */
2169static int find_probe_trace_events_from_map(struct perf_probe_event *pev,
2170 struct probe_trace_event **tevs,
2171 int max_tevs, const char *target)
2172{
2173 struct map *map = NULL;
2174 struct kmap *kmap = NULL;
2175 struct ref_reloc_sym *reloc_sym = NULL;
2015 struct symbol *sym; 2176 struct symbol *sym;
2016 int ret, i; 2177 struct rb_node *nd;
2017 struct probe_trace_event *tev; 2178 struct probe_trace_event *tev;
2179 struct perf_probe_point *pp = &pev->point;
2180 struct probe_trace_point *tp;
2181 int ret, i;
2018 2182
2019 if (pev->uprobes && !pev->group) { 2183 /* Init maps of given executable or kernel */
2020 /* Replace group name if not given */ 2184 if (pev->uprobes)
2021 ret = convert_exec_to_group(target, &pev->group); 2185 map = dso__new_map(target);
2022 if (ret != 0) { 2186 else
2023 pr_warning("Failed to make a group name.\n"); 2187 map = kernel_get_module_map(target);
2024 return ret; 2188 if (!map) {
2025 } 2189 ret = -EINVAL;
2190 goto out;
2026 } 2191 }
2027 2192
2028 /* Convert perf_probe_event with debuginfo */ 2193 /*
2029 ret = try_to_find_probe_trace_events(pev, tevs, max_tevs, target); 2194 * Load matched symbols: Since the different local symbols may have
2030 if (ret != 0) 2195 * same name but different addresses, this lists all the symbols.
2031 return ret; /* Found in debuginfo or got an error */ 2196 */
2032 2197 num_matched_functions = 0;
2033 if (pev->uprobes) { 2198 looking_function_name = pp->function;
2034 ret = convert_name_to_addr(pev, target); 2199 ret = map__load(map, probe_function_filter);
2035 if (ret < 0) 2200 if (ret || num_matched_functions == 0) {
2036 return ret; 2201 pr_err("Failed to find symbol %s in %s\n", pp->function,
2202 target ? : "kernel");
2203 ret = -ENOENT;
2204 goto out;
2205 } else if (num_matched_functions > max_tevs) {
2206 pr_err("Too many functions matched in %s\n",
2207 target ? : "kernel");
2208 ret = -E2BIG;
2209 goto out;
2037 } 2210 }
2038 2211
2039 /* Allocate trace event buffer */ 2212 if (!pev->uprobes) {
2040 tev = *tevs = zalloc(sizeof(struct probe_trace_event)); 2213 kmap = map__kmap(map);
2041 if (tev == NULL) 2214 reloc_sym = kmap->ref_reloc_sym;
2042 return -ENOMEM; 2215 if (!reloc_sym) {
2216 pr_warning("Relocated base symbol is not found!\n");
2217 ret = -EINVAL;
2218 goto out;
2219 }
2220 }
2043 2221
2044 /* Copy parameters */ 2222 /* Setup result trace-probe-events */
2045 tev->point.symbol = strdup(pev->point.function); 2223 *tevs = zalloc(sizeof(*tev) * num_matched_functions);
2046 if (tev->point.symbol == NULL) { 2224 if (!*tevs) {
2047 ret = -ENOMEM; 2225 ret = -ENOMEM;
2048 goto error; 2226 goto out;
2049 } 2227 }
2050 2228
2051 if (target) { 2229 ret = 0;
2052 tev->point.module = strdup(target); 2230 map__for_each_symbol(map, sym, nd) {
2053 if (tev->point.module == NULL) { 2231 tev = (*tevs) + ret;
2054 ret = -ENOMEM; 2232 tp = &tev->point;
2055 goto error; 2233 if (ret == num_matched_functions) {
2234 pr_warning("Too many symbols are listed. Skip it.\n");
2235 break;
2056 } 2236 }
2057 } 2237 ret++;
2058
2059 tev->point.offset = pev->point.offset;
2060 tev->point.retprobe = pev->point.retprobe;
2061 tev->nargs = pev->nargs;
2062 tev->uprobes = pev->uprobes;
2063 2238
2064 if (tev->nargs) { 2239 if (pp->offset > sym->end - sym->start) {
2065 tev->args = zalloc(sizeof(struct probe_trace_arg) 2240 pr_warning("Offset %ld is bigger than the size of %s\n",
2066 * tev->nargs); 2241 pp->offset, sym->name);
2067 if (tev->args == NULL) { 2242 ret = -ENOENT;
2068 ret = -ENOMEM; 2243 goto err_out;
2069 goto error; 2244 }
2245 /* Add one probe point */
2246 tp->address = map->unmap_ip(map, sym->start) + pp->offset;
2247 if (reloc_sym) {
2248 tp->symbol = strdup_or_goto(reloc_sym->name, nomem_out);
2249 tp->offset = tp->address - reloc_sym->addr;
2250 } else {
2251 tp->symbol = strdup_or_goto(sym->name, nomem_out);
2252 tp->offset = pp->offset;
2253 }
2254 tp->retprobe = pp->retprobe;
2255 if (target)
2256 tev->point.module = strdup_or_goto(target, nomem_out);
2257 tev->uprobes = pev->uprobes;
2258 tev->nargs = pev->nargs;
2259 if (tev->nargs) {
2260 tev->args = zalloc(sizeof(struct probe_trace_arg) *
2261 tev->nargs);
2262 if (tev->args == NULL)
2263 goto nomem_out;
2070 } 2264 }
2071 for (i = 0; i < tev->nargs; i++) { 2265 for (i = 0; i < tev->nargs; i++) {
2072 if (pev->args[i].name) { 2266 if (pev->args[i].name)
2073 tev->args[i].name = strdup(pev->args[i].name); 2267 tev->args[i].name =
2074 if (tev->args[i].name == NULL) { 2268 strdup_or_goto(pev->args[i].name,
2075 ret = -ENOMEM; 2269 nomem_out);
2076 goto error; 2270
2077 } 2271 tev->args[i].value = strdup_or_goto(pev->args[i].var,
2078 } 2272 nomem_out);
2079 tev->args[i].value = strdup(pev->args[i].var); 2273 if (pev->args[i].type)
2080 if (tev->args[i].value == NULL) { 2274 tev->args[i].type =
2081 ret = -ENOMEM; 2275 strdup_or_goto(pev->args[i].type,
2082 goto error; 2276 nomem_out);
2083 }
2084 if (pev->args[i].type) {
2085 tev->args[i].type = strdup(pev->args[i].type);
2086 if (tev->args[i].type == NULL) {
2087 ret = -ENOMEM;
2088 goto error;
2089 }
2090 }
2091 } 2277 }
2092 } 2278 }
2093 2279
2094 if (pev->uprobes) 2280out:
2095 return 1; 2281 if (map && pev->uprobes) {
2282 /* Only when using uprobe(exec) map needs to be released */
2283 dso__delete(map->dso);
2284 map__delete(map);
2285 }
2286 return ret;
2096 2287
2097 /* Currently just checking function name from symbol map */ 2288nomem_out:
2098 sym = __find_kernel_function_by_name(tev->point.symbol, NULL); 2289 ret = -ENOMEM;
2099 if (!sym) { 2290err_out:
2100 pr_warning("Kernel symbol \'%s\' not found.\n", 2291 clear_probe_trace_events(*tevs, num_matched_functions);
2101 tev->point.symbol); 2292 zfree(tevs);
2102 ret = -ENOENT; 2293 goto out;
2103 goto error; 2294}
2104 } else if (tev->point.offset > sym->end - sym->start) { 2295
2105 pr_warning("Offset specified is greater than size of %s\n", 2296static int convert_to_probe_trace_events(struct perf_probe_event *pev,
2106 tev->point.symbol); 2297 struct probe_trace_event **tevs,
2107 ret = -ENOENT; 2298 int max_tevs, const char *target)
2108 goto error; 2299{
2300 int ret;
2109 2301
2302 if (pev->uprobes && !pev->group) {
2303 /* Replace group name if not given */
2304 ret = convert_exec_to_group(target, &pev->group);
2305 if (ret != 0) {
2306 pr_warning("Failed to make a group name.\n");
2307 return ret;
2308 }
2110 } 2309 }
2111 2310
2112 return 1; 2311 /* Convert perf_probe_event with debuginfo */
2113error: 2312 ret = try_to_find_probe_trace_events(pev, tevs, max_tevs, target);
2114 clear_probe_trace_event(tev); 2313 if (ret != 0)
2115 free(tev); 2314 return ret; /* Found in debuginfo or got an error */
2116 *tevs = NULL; 2315
2117 return ret; 2316 return find_probe_trace_events_from_map(pev, tevs, max_tevs, target);
2118} 2317}
2119 2318
2120struct __event_package { 2319struct __event_package {
@@ -2135,12 +2334,7 @@ int add_perf_probe_events(struct perf_probe_event *pevs, int npevs,
2135 if (pkgs == NULL) 2334 if (pkgs == NULL)
2136 return -ENOMEM; 2335 return -ENOMEM;
2137 2336
2138 if (!pevs->uprobes) 2337 ret = init_symbol_maps(pevs->uprobes);
2139 /* Init vmlinux path */
2140 ret = init_vmlinux();
2141 else
2142 ret = init_user_exec();
2143
2144 if (ret < 0) { 2338 if (ret < 0) {
2145 free(pkgs); 2339 free(pkgs);
2146 return ret; 2340 return ret;
@@ -2174,6 +2368,7 @@ end:
2174 zfree(&pkgs[i].tevs); 2368 zfree(&pkgs[i].tevs);
2175 } 2369 }
2176 free(pkgs); 2370 free(pkgs);
2371 exit_symbol_maps();
2177 2372
2178 return ret; 2373 return ret;
2179} 2374}
@@ -2323,159 +2518,51 @@ static struct strfilter *available_func_filter;
2323static int filter_available_functions(struct map *map __maybe_unused, 2518static int filter_available_functions(struct map *map __maybe_unused,
2324 struct symbol *sym) 2519 struct symbol *sym)
2325{ 2520{
2326 if (sym->binding == STB_GLOBAL && 2521 if ((sym->binding == STB_GLOBAL || sym->binding == STB_LOCAL) &&
2327 strfilter__compare(available_func_filter, sym->name)) 2522 strfilter__compare(available_func_filter, sym->name))
2328 return 0; 2523 return 0;
2329 return 1; 2524 return 1;
2330} 2525}
2331 2526
2332static int __show_available_funcs(struct map *map) 2527int show_available_funcs(const char *target, struct strfilter *_filter,
2333{ 2528 bool user)
2334 if (map__load(map, filter_available_functions)) {
2335 pr_err("Failed to load map.\n");
2336 return -EINVAL;
2337 }
2338 if (!dso__sorted_by_name(map->dso, map->type))
2339 dso__sort_by_name(map->dso, map->type);
2340
2341 dso__fprintf_symbols_by_name(map->dso, map->type, stdout);
2342 return 0;
2343}
2344
2345static int available_kernel_funcs(const char *module)
2346{ 2529{
2347 struct map *map; 2530 struct map *map;
2348 int ret; 2531 int ret;
2349 2532
2350 ret = init_vmlinux(); 2533 ret = init_symbol_maps(user);
2351 if (ret < 0) 2534 if (ret < 0)
2352 return ret; 2535 return ret;
2353 2536
2354 map = kernel_get_module_map(module); 2537 /* Get a symbol map */
2538 if (user)
2539 map = dso__new_map(target);
2540 else
2541 map = kernel_get_module_map(target);
2355 if (!map) { 2542 if (!map) {
2356 pr_err("Failed to find %s map.\n", (module) ? : "kernel"); 2543 pr_err("Failed to get a map for %s\n", (target) ? : "kernel");
2357 return -EINVAL; 2544 return -EINVAL;
2358 } 2545 }
2359 return __show_available_funcs(map);
2360}
2361
2362static int available_user_funcs(const char *target)
2363{
2364 struct map *map;
2365 int ret;
2366
2367 ret = init_user_exec();
2368 if (ret < 0)
2369 return ret;
2370
2371 map = dso__new_map(target);
2372 ret = __show_available_funcs(map);
2373 dso__delete(map->dso);
2374 map__delete(map);
2375 return ret;
2376}
2377 2546
2378int show_available_funcs(const char *target, struct strfilter *_filter, 2547 /* Load symbols with given filter */
2379 bool user)
2380{
2381 setup_pager();
2382 available_func_filter = _filter; 2548 available_func_filter = _filter;
2383
2384 if (!user)
2385 return available_kernel_funcs(target);
2386
2387 return available_user_funcs(target);
2388}
2389
2390/*
2391 * uprobe_events only accepts address:
2392 * Convert function and any offset to address
2393 */
2394static int convert_name_to_addr(struct perf_probe_event *pev, const char *exec)
2395{
2396 struct perf_probe_point *pp = &pev->point;
2397 struct symbol *sym;
2398 struct map *map = NULL;
2399 char *function = NULL;
2400 int ret = -EINVAL;
2401 unsigned long long vaddr = 0;
2402
2403 if (!pp->function) {
2404 pr_warning("No function specified for uprobes");
2405 goto out;
2406 }
2407
2408 function = strdup(pp->function);
2409 if (!function) {
2410 pr_warning("Failed to allocate memory by strdup.\n");
2411 ret = -ENOMEM;
2412 goto out;
2413 }
2414
2415 map = dso__new_map(exec);
2416 if (!map) {
2417 pr_warning("Cannot find appropriate DSO for %s.\n", exec);
2418 goto out;
2419 }
2420 available_func_filter = strfilter__new(function, NULL);
2421 if (map__load(map, filter_available_functions)) { 2549 if (map__load(map, filter_available_functions)) {
2422 pr_err("Failed to load map.\n"); 2550 pr_err("Failed to load symbols in %s\n", (target) ? : "kernel");
2423 goto out; 2551 goto end;
2424 }
2425
2426 sym = map__find_symbol_by_name(map, function, NULL);
2427 if (!sym) {
2428 pr_warning("Cannot find %s in DSO %s\n", function, exec);
2429 goto out;
2430 }
2431
2432 if (map->start > sym->start)
2433 vaddr = map->start;
2434 vaddr += sym->start + pp->offset + map->pgoff;
2435 pp->offset = 0;
2436
2437 if (!pev->event) {
2438 pev->event = function;
2439 function = NULL;
2440 }
2441 if (!pev->group) {
2442 char *ptr1, *ptr2, *exec_copy;
2443
2444 pev->group = zalloc(sizeof(char *) * 64);
2445 exec_copy = strdup(exec);
2446 if (!exec_copy) {
2447 ret = -ENOMEM;
2448 pr_warning("Failed to copy exec string.\n");
2449 goto out;
2450 }
2451
2452 ptr1 = strdup(basename(exec_copy));
2453 if (ptr1) {
2454 ptr2 = strpbrk(ptr1, "-._");
2455 if (ptr2)
2456 *ptr2 = '\0';
2457 e_snprintf(pev->group, 64, "%s_%s", PERFPROBE_GROUP,
2458 ptr1);
2459 free(ptr1);
2460 }
2461 free(exec_copy);
2462 }
2463 free(pp->function);
2464 pp->function = zalloc(sizeof(char *) * MAX_PROBE_ARGS);
2465 if (!pp->function) {
2466 ret = -ENOMEM;
2467 pr_warning("Failed to allocate memory by zalloc.\n");
2468 goto out;
2469 } 2552 }
2470 e_snprintf(pp->function, MAX_PROBE_ARGS, "0x%llx", vaddr); 2553 if (!dso__sorted_by_name(map->dso, map->type))
2471 ret = 0; 2554 dso__sort_by_name(map->dso, map->type);
2472 2555
2473out: 2556 /* Show all (filtered) symbols */
2474 if (map) { 2557 setup_pager();
2558 dso__fprintf_symbols_by_name(map->dso, map->type, stdout);
2559end:
2560 if (user) {
2475 dso__delete(map->dso); 2561 dso__delete(map->dso);
2476 map__delete(map); 2562 map__delete(map);
2477 } 2563 }
2478 if (function) 2564 exit_symbol_maps();
2479 free(function); 2565
2480 return ret; 2566 return ret;
2481} 2567}
2568
diff --git a/tools/perf/util/probe-event.h b/tools/perf/util/probe-event.h
index fcaf7273e85a..776c9347a3b6 100644
--- a/tools/perf/util/probe-event.h
+++ b/tools/perf/util/probe-event.h
@@ -2,6 +2,7 @@
2#define _PROBE_EVENT_H 2#define _PROBE_EVENT_H
3 3
4#include <stdbool.h> 4#include <stdbool.h>
5#include "intlist.h"
5#include "strlist.h" 6#include "strlist.h"
6#include "strfilter.h" 7#include "strfilter.h"
7 8
@@ -76,13 +77,6 @@ struct perf_probe_event {
76 struct perf_probe_arg *args; /* Arguments */ 77 struct perf_probe_arg *args; /* Arguments */
77}; 78};
78 79
79
80/* Line number container */
81struct line_node {
82 struct list_head list;
83 int line;
84};
85
86/* Line range */ 80/* Line range */
87struct line_range { 81struct line_range {
88 char *file; /* File name */ 82 char *file; /* File name */
@@ -92,7 +86,7 @@ struct line_range {
92 int offset; /* Start line offset */ 86 int offset; /* Start line offset */
93 char *path; /* Real path name */ 87 char *path; /* Real path name */
94 char *comp_dir; /* Compile directory */ 88 char *comp_dir; /* Compile directory */
95 struct list_head line_list; /* Visible lines */ 89 struct intlist *line_list; /* Visible lines */
96}; 90};
97 91
98/* List of variables */ 92/* List of variables */
@@ -124,7 +118,7 @@ extern int parse_line_range_desc(const char *cmd, struct line_range *lr);
124extern void line_range__clear(struct line_range *lr); 118extern void line_range__clear(struct line_range *lr);
125 119
126/* Initialize line range */ 120/* Initialize line range */
127extern void line_range__init(struct line_range *lr); 121extern int line_range__init(struct line_range *lr);
128 122
129/* Internal use: Return kernel/module path */ 123/* Internal use: Return kernel/module path */
130extern const char *kernel_get_module_path(const char *module); 124extern const char *kernel_get_module_path(const char *module);
diff --git a/tools/perf/util/probe-finder.c b/tools/perf/util/probe-finder.c
index 061edb162b5b..df0238654698 100644
--- a/tools/perf/util/probe-finder.c
+++ b/tools/perf/util/probe-finder.c
@@ -34,7 +34,9 @@
34 34
35#include <linux/bitops.h> 35#include <linux/bitops.h>
36#include "event.h" 36#include "event.h"
37#include "dso.h"
37#include "debug.h" 38#include "debug.h"
39#include "intlist.h"
38#include "util.h" 40#include "util.h"
39#include "symbol.h" 41#include "symbol.h"
40#include "probe-finder.h" 42#include "probe-finder.h"
@@ -42,65 +44,6 @@
42/* Kprobe tracer basic type is up to u64 */ 44/* Kprobe tracer basic type is up to u64 */
43#define MAX_BASIC_TYPE_BITS 64 45#define MAX_BASIC_TYPE_BITS 64
44 46
45/* Line number list operations */
46
47/* Add a line to line number list */
48static int line_list__add_line(struct list_head *head, int line)
49{
50 struct line_node *ln;
51 struct list_head *p;
52
53 /* Reverse search, because new line will be the last one */
54 list_for_each_entry_reverse(ln, head, list) {
55 if (ln->line < line) {
56 p = &ln->list;
57 goto found;
58 } else if (ln->line == line) /* Already exist */
59 return 1;
60 }
61 /* List is empty, or the smallest entry */
62 p = head;
63found:
64 pr_debug("line list: add a line %u\n", line);
65 ln = zalloc(sizeof(struct line_node));
66 if (ln == NULL)
67 return -ENOMEM;
68 ln->line = line;
69 INIT_LIST_HEAD(&ln->list);
70 list_add(&ln->list, p);
71 return 0;
72}
73
74/* Check if the line in line number list */
75static int line_list__has_line(struct list_head *head, int line)
76{
77 struct line_node *ln;
78
79 /* Reverse search, because new line will be the last one */
80 list_for_each_entry(ln, head, list)
81 if (ln->line == line)
82 return 1;
83
84 return 0;
85}
86
87/* Init line number list */
88static void line_list__init(struct list_head *head)
89{
90 INIT_LIST_HEAD(head);
91}
92
93/* Free line number list */
94static void line_list__free(struct list_head *head)
95{
96 struct line_node *ln;
97 while (!list_empty(head)) {
98 ln = list_first_entry(head, struct line_node, list);
99 list_del(&ln->list);
100 free(ln);
101 }
102}
103
104/* Dwarf FL wrappers */ 47/* Dwarf FL wrappers */
105static char *debuginfo_path; /* Currently dummy */ 48static char *debuginfo_path; /* Currently dummy */
106 49
@@ -147,80 +90,7 @@ error:
147 return -ENOENT; 90 return -ENOENT;
148} 91}
149 92
150#if _ELFUTILS_PREREQ(0, 148) 93static struct debuginfo *__debuginfo__new(const char *path)
151/* This method is buggy if elfutils is older than 0.148 */
152static int __linux_kernel_find_elf(Dwfl_Module *mod,
153 void **userdata,
154 const char *module_name,
155 Dwarf_Addr base,
156 char **file_name, Elf **elfp)
157{
158 int fd;
159 const char *path = kernel_get_module_path(module_name);
160
161 pr_debug2("Use file %s for %s\n", path, module_name);
162 if (path) {
163 fd = open(path, O_RDONLY);
164 if (fd >= 0) {
165 *file_name = strdup(path);
166 return fd;
167 }
168 }
169 /* If failed, try to call standard method */
170 return dwfl_linux_kernel_find_elf(mod, userdata, module_name, base,
171 file_name, elfp);
172}
173
174static const Dwfl_Callbacks kernel_callbacks = {
175 .find_debuginfo = dwfl_standard_find_debuginfo,
176 .debuginfo_path = &debuginfo_path,
177
178 .find_elf = __linux_kernel_find_elf,
179 .section_address = dwfl_linux_kernel_module_section_address,
180};
181
182/* Get a Dwarf from live kernel image */
183static int debuginfo__init_online_kernel_dwarf(struct debuginfo *dbg,
184 Dwarf_Addr addr)
185{
186 dbg->dwfl = dwfl_begin(&kernel_callbacks);
187 if (!dbg->dwfl)
188 return -EINVAL;
189
190 /* Load the kernel dwarves: Don't care the result here */
191 dwfl_linux_kernel_report_kernel(dbg->dwfl);
192 dwfl_linux_kernel_report_modules(dbg->dwfl);
193
194 dbg->dbg = dwfl_addrdwarf(dbg->dwfl, addr, &dbg->bias);
195 /* Here, check whether we could get a real dwarf */
196 if (!dbg->dbg) {
197 pr_debug("Failed to find kernel dwarf at %lx\n",
198 (unsigned long)addr);
199 dwfl_end(dbg->dwfl);
200 memset(dbg, 0, sizeof(*dbg));
201 return -ENOENT;
202 }
203
204 return 0;
205}
206#else
207/* With older elfutils, this just support kernel module... */
208static int debuginfo__init_online_kernel_dwarf(struct debuginfo *dbg,
209 Dwarf_Addr addr __maybe_unused)
210{
211 const char *path = kernel_get_module_path("kernel");
212
213 if (!path) {
214 pr_err("Failed to find vmlinux path\n");
215 return -ENOENT;
216 }
217
218 pr_debug2("Use file %s for debuginfo\n", path);
219 return debuginfo__init_offline_dwarf(dbg, path);
220}
221#endif
222
223struct debuginfo *debuginfo__new(const char *path)
224{ 94{
225 struct debuginfo *dbg = zalloc(sizeof(*dbg)); 95 struct debuginfo *dbg = zalloc(sizeof(*dbg));
226 if (!dbg) 96 if (!dbg)
@@ -228,21 +98,44 @@ struct debuginfo *debuginfo__new(const char *path)
228 98
229 if (debuginfo__init_offline_dwarf(dbg, path) < 0) 99 if (debuginfo__init_offline_dwarf(dbg, path) < 0)
230 zfree(&dbg); 100 zfree(&dbg);
231 101 if (dbg)
102 pr_debug("Open Debuginfo file: %s\n", path);
232 return dbg; 103 return dbg;
233} 104}
234 105
235struct debuginfo *debuginfo__new_online_kernel(unsigned long addr) 106enum dso_binary_type distro_dwarf_types[] = {
236{ 107 DSO_BINARY_TYPE__FEDORA_DEBUGINFO,
237 struct debuginfo *dbg = zalloc(sizeof(*dbg)); 108 DSO_BINARY_TYPE__UBUNTU_DEBUGINFO,
109 DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO,
110 DSO_BINARY_TYPE__BUILDID_DEBUGINFO,
111 DSO_BINARY_TYPE__NOT_FOUND,
112};
238 113
239 if (!dbg) 114struct debuginfo *debuginfo__new(const char *path)
240 return NULL; 115{
116 enum dso_binary_type *type;
117 char buf[PATH_MAX], nil = '\0';
118 struct dso *dso;
119 struct debuginfo *dinfo = NULL;
120
121 /* Try to open distro debuginfo files */
122 dso = dso__new(path);
123 if (!dso)
124 goto out;
241 125
242 if (debuginfo__init_online_kernel_dwarf(dbg, (Dwarf_Addr)addr) < 0) 126 for (type = distro_dwarf_types;
243 zfree(&dbg); 127 !dinfo && *type != DSO_BINARY_TYPE__NOT_FOUND;
128 type++) {
129 if (dso__read_binary_type_filename(dso, *type, &nil,
130 buf, PATH_MAX) < 0)
131 continue;
132 dinfo = __debuginfo__new(buf);
133 }
134 dso__delete(dso);
244 135
245 return dbg; 136out:
137 /* if failed to open all distro debuginfo, open given binary */
138 return dinfo ? : __debuginfo__new(path);
246} 139}
247 140
248void debuginfo__delete(struct debuginfo *dbg) 141void debuginfo__delete(struct debuginfo *dbg)
@@ -880,7 +773,7 @@ static int find_probe_point_by_line(struct probe_finder *pf)
880} 773}
881 774
882/* Find lines which match lazy pattern */ 775/* Find lines which match lazy pattern */
883static int find_lazy_match_lines(struct list_head *head, 776static int find_lazy_match_lines(struct intlist *list,
884 const char *fname, const char *pat) 777 const char *fname, const char *pat)
885{ 778{
886 FILE *fp; 779 FILE *fp;
@@ -901,7 +794,7 @@ static int find_lazy_match_lines(struct list_head *head,
901 line[len - 1] = '\0'; 794 line[len - 1] = '\0';
902 795
903 if (strlazymatch(line, pat)) { 796 if (strlazymatch(line, pat)) {
904 line_list__add_line(head, linenum); 797 intlist__add(list, linenum);
905 count++; 798 count++;
906 } 799 }
907 linenum++; 800 linenum++;
@@ -924,7 +817,7 @@ static int probe_point_lazy_walker(const char *fname, int lineno,
924 Dwarf_Die *sc_die, die_mem; 817 Dwarf_Die *sc_die, die_mem;
925 int ret; 818 int ret;
926 819
927 if (!line_list__has_line(&pf->lcache, lineno) || 820 if (!intlist__has_entry(pf->lcache, lineno) ||
928 strtailcmp(fname, pf->fname) != 0) 821 strtailcmp(fname, pf->fname) != 0)
929 return 0; 822 return 0;
930 823
@@ -952,9 +845,9 @@ static int find_probe_point_lazy(Dwarf_Die *sp_die, struct probe_finder *pf)
952{ 845{
953 int ret = 0; 846 int ret = 0;
954 847
955 if (list_empty(&pf->lcache)) { 848 if (intlist__empty(pf->lcache)) {
956 /* Matching lazy line pattern */ 849 /* Matching lazy line pattern */
957 ret = find_lazy_match_lines(&pf->lcache, pf->fname, 850 ret = find_lazy_match_lines(pf->lcache, pf->fname,
958 pf->pev->point.lazy_line); 851 pf->pev->point.lazy_line);
959 if (ret <= 0) 852 if (ret <= 0)
960 return ret; 853 return ret;
@@ -1096,7 +989,9 @@ static int debuginfo__find_probes(struct debuginfo *dbg,
1096#endif 989#endif
1097 990
1098 off = 0; 991 off = 0;
1099 line_list__init(&pf->lcache); 992 pf->lcache = intlist__new(NULL);
993 if (!pf->lcache)
994 return -ENOMEM;
1100 995
1101 /* Fastpath: lookup by function name from .debug_pubnames section */ 996 /* Fastpath: lookup by function name from .debug_pubnames section */
1102 if (pp->function) { 997 if (pp->function) {
@@ -1149,7 +1044,8 @@ static int debuginfo__find_probes(struct debuginfo *dbg,
1149 } 1044 }
1150 1045
1151found: 1046found:
1152 line_list__free(&pf->lcache); 1047 intlist__delete(pf->lcache);
1048 pf->lcache = NULL;
1153 1049
1154 return ret; 1050 return ret;
1155} 1051}
@@ -1537,7 +1433,7 @@ static int line_range_add_line(const char *src, unsigned int lineno,
1537 if (lr->path == NULL) 1433 if (lr->path == NULL)
1538 return -ENOMEM; 1434 return -ENOMEM;
1539 } 1435 }
1540 return line_list__add_line(&lr->line_list, lineno); 1436 return intlist__add(lr->line_list, lineno);
1541} 1437}
1542 1438
1543static int line_range_walk_cb(const char *fname, int lineno, 1439static int line_range_walk_cb(const char *fname, int lineno,
@@ -1565,7 +1461,7 @@ static int find_line_range_by_line(Dwarf_Die *sp_die, struct line_finder *lf)
1565 1461
1566 /* Update status */ 1462 /* Update status */
1567 if (ret >= 0) 1463 if (ret >= 0)
1568 if (!list_empty(&lf->lr->line_list)) 1464 if (!intlist__empty(lf->lr->line_list))
1569 ret = lf->found = 1; 1465 ret = lf->found = 1;
1570 else 1466 else
1571 ret = 0; /* Lines are not found */ 1467 ret = 0; /* Lines are not found */
diff --git a/tools/perf/util/probe-finder.h b/tools/perf/util/probe-finder.h
index ffc33cdd25cc..92590b2c7e1c 100644
--- a/tools/perf/util/probe-finder.h
+++ b/tools/perf/util/probe-finder.h
@@ -3,6 +3,7 @@
3 3
4#include <stdbool.h> 4#include <stdbool.h>
5#include "util.h" 5#include "util.h"
6#include "intlist.h"
6#include "probe-event.h" 7#include "probe-event.h"
7 8
8#define MAX_PROBE_BUFFER 1024 9#define MAX_PROBE_BUFFER 1024
@@ -29,8 +30,8 @@ struct debuginfo {
29 Dwarf_Addr bias; 30 Dwarf_Addr bias;
30}; 31};
31 32
33/* This also tries to open distro debuginfo */
32extern struct debuginfo *debuginfo__new(const char *path); 34extern struct debuginfo *debuginfo__new(const char *path);
33extern struct debuginfo *debuginfo__new_online_kernel(unsigned long addr);
34extern void debuginfo__delete(struct debuginfo *dbg); 35extern void debuginfo__delete(struct debuginfo *dbg);
35 36
36/* Find probe_trace_events specified by perf_probe_event from debuginfo */ 37/* Find probe_trace_events specified by perf_probe_event from debuginfo */
@@ -66,7 +67,7 @@ struct probe_finder {
66 const char *fname; /* Real file name */ 67 const char *fname; /* Real file name */
67 Dwarf_Die cu_die; /* Current CU */ 68 Dwarf_Die cu_die; /* Current CU */
68 Dwarf_Die sp_die; 69 Dwarf_Die sp_die;
69 struct list_head lcache; /* Line cache for lazy match */ 70 struct intlist *lcache; /* Line cache for lazy match */
70 71
71 /* For variable searching */ 72 /* For variable searching */
72#if _ELFUTILS_PREREQ(0, 142) 73#if _ELFUTILS_PREREQ(0, 142)
diff --git a/tools/perf/util/python-ext-sources b/tools/perf/util/python-ext-sources
index 595bfc73d2ed..16a475a7d492 100644
--- a/tools/perf/util/python-ext-sources
+++ b/tools/perf/util/python-ext-sources
@@ -17,6 +17,6 @@ util/xyarray.c
17util/cgroup.c 17util/cgroup.c
18util/rblist.c 18util/rblist.c
19util/strlist.c 19util/strlist.c
20util/fs.c 20../lib/api/fs/fs.c
21util/trace-event.c 21util/trace-event.c
22../../lib/rbtree.c 22../../lib/rbtree.c
diff --git a/tools/perf/util/record.c b/tools/perf/util/record.c
index 373762501dad..049e0a09ccd3 100644
--- a/tools/perf/util/record.c
+++ b/tools/perf/util/record.c
@@ -2,7 +2,7 @@
2#include "evsel.h" 2#include "evsel.h"
3#include "cpumap.h" 3#include "cpumap.h"
4#include "parse-events.h" 4#include "parse-events.h"
5#include "fs.h" 5#include <api/fs/fs.h>
6#include "util.h" 6#include "util.h"
7 7
8typedef void (*setup_probe_fn_t)(struct perf_evsel *evsel); 8typedef void (*setup_probe_fn_t)(struct perf_evsel *evsel);
diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c
index 5da6ce74c676..55960f22233c 100644
--- a/tools/perf/util/session.c
+++ b/tools/perf/util/session.c
@@ -702,11 +702,12 @@ static void regs_dump__printf(u64 mask, u64 *regs)
702 } 702 }
703} 703}
704 704
705static void regs_user__printf(struct perf_sample *sample, u64 mask) 705static void regs_user__printf(struct perf_sample *sample)
706{ 706{
707 struct regs_dump *user_regs = &sample->user_regs; 707 struct regs_dump *user_regs = &sample->user_regs;
708 708
709 if (user_regs->regs) { 709 if (user_regs->regs) {
710 u64 mask = user_regs->mask;
710 printf("... user regs: mask 0x%" PRIx64 "\n", mask); 711 printf("... user regs: mask 0x%" PRIx64 "\n", mask);
711 regs_dump__printf(mask, user_regs->regs); 712 regs_dump__printf(mask, user_regs->regs);
712 } 713 }
@@ -793,7 +794,7 @@ static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
793 if (!dump_trace) 794 if (!dump_trace)
794 return; 795 return;
795 796
796 printf("(IP, %d): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 797 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
797 event->header.misc, sample->pid, sample->tid, sample->ip, 798 event->header.misc, sample->pid, sample->tid, sample->ip,
798 sample->period, sample->addr); 799 sample->period, sample->addr);
799 800
@@ -806,7 +807,7 @@ static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
806 branch_stack__printf(sample); 807 branch_stack__printf(sample);
807 808
808 if (sample_type & PERF_SAMPLE_REGS_USER) 809 if (sample_type & PERF_SAMPLE_REGS_USER)
809 regs_user__printf(sample, evsel->attr.sample_regs_user); 810 regs_user__printf(sample);
810 811
811 if (sample_type & PERF_SAMPLE_STACK_USER) 812 if (sample_type & PERF_SAMPLE_STACK_USER)
812 stack_user__printf(&sample->user_stack); 813 stack_user__printf(&sample->user_stack);
diff --git a/tools/perf/util/symbol-elf.c b/tools/perf/util/symbol-elf.c
index 516d19fb999b..3b7dbf51d4a9 100644
--- a/tools/perf/util/symbol-elf.c
+++ b/tools/perf/util/symbol-elf.c
@@ -506,6 +506,8 @@ int filename__read_debuglink(const char *filename, char *debuglink,
506 /* the start of this section is a zero-terminated string */ 506 /* the start of this section is a zero-terminated string */
507 strncpy(debuglink, data->d_buf, size); 507 strncpy(debuglink, data->d_buf, size);
508 508
509 err = 0;
510
509out_elf_end: 511out_elf_end:
510 elf_end(elf); 512 elf_end(elf);
511out_close: 513out_close:
diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index e89afc097d8a..95e249779931 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -410,7 +410,7 @@ struct symbol *dso__find_symbol(struct dso *dso,
410 return symbols__find(&dso->symbols[type], addr); 410 return symbols__find(&dso->symbols[type], addr);
411} 411}
412 412
413struct symbol *dso__first_symbol(struct dso *dso, enum map_type type) 413static struct symbol *dso__first_symbol(struct dso *dso, enum map_type type)
414{ 414{
415 return symbols__first(&dso->symbols[type]); 415 return symbols__first(&dso->symbols[type]);
416} 416}
@@ -1251,6 +1251,46 @@ out_failure:
1251 return -1; 1251 return -1;
1252} 1252}
1253 1253
1254static bool dso__is_compatible_symtab_type(struct dso *dso, bool kmod,
1255 enum dso_binary_type type)
1256{
1257 switch (type) {
1258 case DSO_BINARY_TYPE__JAVA_JIT:
1259 case DSO_BINARY_TYPE__DEBUGLINK:
1260 case DSO_BINARY_TYPE__SYSTEM_PATH_DSO:
1261 case DSO_BINARY_TYPE__FEDORA_DEBUGINFO:
1262 case DSO_BINARY_TYPE__UBUNTU_DEBUGINFO:
1263 case DSO_BINARY_TYPE__BUILDID_DEBUGINFO:
1264 case DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO:
1265 return !kmod && dso->kernel == DSO_TYPE_USER;
1266
1267 case DSO_BINARY_TYPE__KALLSYMS:
1268 case DSO_BINARY_TYPE__VMLINUX:
1269 case DSO_BINARY_TYPE__KCORE:
1270 return dso->kernel == DSO_TYPE_KERNEL;
1271
1272 case DSO_BINARY_TYPE__GUEST_KALLSYMS:
1273 case DSO_BINARY_TYPE__GUEST_VMLINUX:
1274 case DSO_BINARY_TYPE__GUEST_KCORE:
1275 return dso->kernel == DSO_TYPE_GUEST_KERNEL;
1276
1277 case DSO_BINARY_TYPE__GUEST_KMODULE:
1278 case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE:
1279 /*
1280 * kernel modules know their symtab type - it's set when
1281 * creating a module dso in machine__new_module().
1282 */
1283 return kmod && dso->symtab_type == type;
1284
1285 case DSO_BINARY_TYPE__BUILD_ID_CACHE:
1286 return true;
1287
1288 case DSO_BINARY_TYPE__NOT_FOUND:
1289 default:
1290 return false;
1291 }
1292}
1293
1254int dso__load(struct dso *dso, struct map *map, symbol_filter_t filter) 1294int dso__load(struct dso *dso, struct map *map, symbol_filter_t filter)
1255{ 1295{
1256 char *name; 1296 char *name;
@@ -1261,6 +1301,7 @@ int dso__load(struct dso *dso, struct map *map, symbol_filter_t filter)
1261 int ss_pos = 0; 1301 int ss_pos = 0;
1262 struct symsrc ss_[2]; 1302 struct symsrc ss_[2];
1263 struct symsrc *syms_ss = NULL, *runtime_ss = NULL; 1303 struct symsrc *syms_ss = NULL, *runtime_ss = NULL;
1304 bool kmod;
1264 1305
1265 dso__set_loaded(dso, map->type); 1306 dso__set_loaded(dso, map->type);
1266 1307
@@ -1301,7 +1342,11 @@ int dso__load(struct dso *dso, struct map *map, symbol_filter_t filter)
1301 if (!name) 1342 if (!name)
1302 return -1; 1343 return -1;
1303 1344
1304 /* Iterate over candidate debug images. 1345 kmod = dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1346 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
1347
1348 /*
1349 * Iterate over candidate debug images.
1305 * Keep track of "interesting" ones (those which have a symtab, dynsym, 1350 * Keep track of "interesting" ones (those which have a symtab, dynsym,
1306 * and/or opd section) for processing. 1351 * and/or opd section) for processing.
1307 */ 1352 */
@@ -1311,6 +1356,9 @@ int dso__load(struct dso *dso, struct map *map, symbol_filter_t filter)
1311 1356
1312 enum dso_binary_type symtab_type = binary_type_symtab[i]; 1357 enum dso_binary_type symtab_type = binary_type_symtab[i];
1313 1358
1359 if (!dso__is_compatible_symtab_type(dso, kmod, symtab_type))
1360 continue;
1361
1314 if (dso__read_binary_type_filename(dso, symtab_type, 1362 if (dso__read_binary_type_filename(dso, symtab_type,
1315 root_dir, name, PATH_MAX)) 1363 root_dir, name, PATH_MAX))
1316 continue; 1364 continue;
@@ -1353,15 +1401,10 @@ int dso__load(struct dso *dso, struct map *map, symbol_filter_t filter)
1353 if (!runtime_ss && syms_ss) 1401 if (!runtime_ss && syms_ss)
1354 runtime_ss = syms_ss; 1402 runtime_ss = syms_ss;
1355 1403
1356 if (syms_ss) { 1404 if (syms_ss)
1357 int km; 1405 ret = dso__load_sym(dso, map, syms_ss, runtime_ss, filter, kmod);
1358 1406 else
1359 km = dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1360 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
1361 ret = dso__load_sym(dso, map, syms_ss, runtime_ss, filter, km);
1362 } else {
1363 ret = -1; 1407 ret = -1;
1364 }
1365 1408
1366 if (ret > 0) { 1409 if (ret > 0) {
1367 int nr_plt; 1410 int nr_plt;
diff --git a/tools/perf/util/symbol.h b/tools/perf/util/symbol.h
index fffe2888a1c7..501e4e722e8e 100644
--- a/tools/perf/util/symbol.h
+++ b/tools/perf/util/symbol.h
@@ -79,6 +79,17 @@ struct symbol {
79void symbol__delete(struct symbol *sym); 79void symbol__delete(struct symbol *sym);
80void symbols__delete(struct rb_root *symbols); 80void symbols__delete(struct rb_root *symbols);
81 81
82/* symbols__for_each_entry - iterate over symbols (rb_root)
83 *
84 * @symbols: the rb_root of symbols
85 * @pos: the 'struct symbol *' to use as a loop cursor
86 * @nd: the 'struct rb_node *' to use as a temporary storage
87 */
88#define symbols__for_each_entry(symbols, pos, nd) \
89 for (nd = rb_first(symbols); \
90 nd && (pos = rb_entry(nd, struct symbol, rb_node)); \
91 nd = rb_next(nd))
92
82static inline size_t symbol__size(const struct symbol *sym) 93static inline size_t symbol__size(const struct symbol *sym)
83{ 94{
84 return sym->end - sym->start + 1; 95 return sym->end - sym->start + 1;
@@ -175,7 +186,7 @@ struct addr_location {
175 struct symbol *sym; 186 struct symbol *sym;
176 u64 addr; 187 u64 addr;
177 char level; 188 char level;
178 bool filtered; 189 u8 filtered;
179 u8 cpumode; 190 u8 cpumode;
180 s32 cpu; 191 s32 cpu;
181}; 192};
@@ -223,7 +234,6 @@ struct symbol *dso__find_symbol(struct dso *dso, enum map_type type,
223 u64 addr); 234 u64 addr);
224struct symbol *dso__find_symbol_by_name(struct dso *dso, enum map_type type, 235struct symbol *dso__find_symbol_by_name(struct dso *dso, enum map_type type,
225 const char *name); 236 const char *name);
226struct symbol *dso__first_symbol(struct dso *dso, enum map_type type);
227 237
228int filename__read_build_id(const char *filename, void *bf, size_t size); 238int filename__read_build_id(const char *filename, void *bf, size_t size);
229int sysfs__read_build_id(const char *filename, void *bf, size_t size); 239int sysfs__read_build_id(const char *filename, void *bf, size_t size);
diff --git a/tools/perf/util/thread.c b/tools/perf/util/thread.c
index 0358882c8910..3ce0498bdae6 100644
--- a/tools/perf/util/thread.c
+++ b/tools/perf/util/thread.c
@@ -142,3 +142,24 @@ int thread__fork(struct thread *thread, struct thread *parent, u64 timestamp)
142 142
143 return 0; 143 return 0;
144} 144}
145
146void thread__find_cpumode_addr_location(struct thread *thread,
147 struct machine *machine,
148 enum map_type type, u64 addr,
149 struct addr_location *al)
150{
151 size_t i;
152 const u8 const cpumodes[] = {
153 PERF_RECORD_MISC_USER,
154 PERF_RECORD_MISC_KERNEL,
155 PERF_RECORD_MISC_GUEST_USER,
156 PERF_RECORD_MISC_GUEST_KERNEL
157 };
158
159 for (i = 0; i < ARRAY_SIZE(cpumodes); i++) {
160 thread__find_addr_location(thread, machine, cpumodes[i], type,
161 addr, al);
162 if (al->map)
163 break;
164 }
165}
diff --git a/tools/perf/util/thread.h b/tools/perf/util/thread.h
index 5b856bf942e1..9b29f085aede 100644
--- a/tools/perf/util/thread.h
+++ b/tools/perf/util/thread.h
@@ -44,12 +44,6 @@ void thread__insert_map(struct thread *thread, struct map *map);
44int thread__fork(struct thread *thread, struct thread *parent, u64 timestamp); 44int thread__fork(struct thread *thread, struct thread *parent, u64 timestamp);
45size_t thread__fprintf(struct thread *thread, FILE *fp); 45size_t thread__fprintf(struct thread *thread, FILE *fp);
46 46
47static inline struct map *thread__find_map(struct thread *thread,
48 enum map_type type, u64 addr)
49{
50 return thread ? map_groups__find(&thread->mg, type, addr) : NULL;
51}
52
53void thread__find_addr_map(struct thread *thread, struct machine *machine, 47void thread__find_addr_map(struct thread *thread, struct machine *machine,
54 u8 cpumode, enum map_type type, u64 addr, 48 u8 cpumode, enum map_type type, u64 addr,
55 struct addr_location *al); 49 struct addr_location *al);
@@ -58,6 +52,11 @@ void thread__find_addr_location(struct thread *thread, struct machine *machine,
58 u8 cpumode, enum map_type type, u64 addr, 52 u8 cpumode, enum map_type type, u64 addr,
59 struct addr_location *al); 53 struct addr_location *al);
60 54
55void thread__find_cpumode_addr_location(struct thread *thread,
56 struct machine *machine,
57 enum map_type type, u64 addr,
58 struct addr_location *al);
59
61static inline void *thread__priv(struct thread *thread) 60static inline void *thread__priv(struct thread *thread)
62{ 61{
63 return thread->priv; 62 return thread->priv;
diff --git a/tools/perf/util/trace-event-parse.c b/tools/perf/util/trace-event-parse.c
index e0d6d07f6848..c36636fd825b 100644
--- a/tools/perf/util/trace-event-parse.c
+++ b/tools/perf/util/trace-event-parse.c
@@ -126,6 +126,7 @@ void event_format__print(struct event_format *event,
126 trace_seq_init(&s); 126 trace_seq_init(&s);
127 pevent_event_info(&s, event, &record); 127 pevent_event_info(&s, event, &record);
128 trace_seq_do_printf(&s); 128 trace_seq_do_printf(&s);
129 trace_seq_destroy(&s);
129} 130}
130 131
131void parse_proc_kallsyms(struct pevent *pevent, 132void parse_proc_kallsyms(struct pevent *pevent,
diff --git a/tools/perf/util/unwind-libdw.c b/tools/perf/util/unwind-libdw.c
new file mode 100644
index 000000000000..67db73ec3dab
--- /dev/null
+++ b/tools/perf/util/unwind-libdw.c
@@ -0,0 +1,210 @@
1#include <linux/compiler.h>
2#include <elfutils/libdw.h>
3#include <elfutils/libdwfl.h>
4#include <inttypes.h>
5#include <errno.h>
6#include "unwind.h"
7#include "unwind-libdw.h"
8#include "machine.h"
9#include "thread.h"
10#include "types.h"
11#include "event.h"
12#include "perf_regs.h"
13
14static char *debuginfo_path;
15
16static const Dwfl_Callbacks offline_callbacks = {
17 .find_debuginfo = dwfl_standard_find_debuginfo,
18 .debuginfo_path = &debuginfo_path,
19 .section_address = dwfl_offline_section_address,
20};
21
22static int __report_module(struct addr_location *al, u64 ip,
23 struct unwind_info *ui)
24{
25 Dwfl_Module *mod;
26 struct dso *dso = NULL;
27
28 thread__find_addr_location(ui->thread, ui->machine,
29 PERF_RECORD_MISC_USER,
30 MAP__FUNCTION, ip, al);
31
32 if (al->map)
33 dso = al->map->dso;
34
35 if (!dso)
36 return 0;
37
38 mod = dwfl_addrmodule(ui->dwfl, ip);
39 if (!mod)
40 mod = dwfl_report_elf(ui->dwfl, dso->short_name,
41 dso->long_name, -1, al->map->start,
42 false);
43
44 return mod && dwfl_addrmodule(ui->dwfl, ip) == mod ? 0 : -1;
45}
46
47static int report_module(u64 ip, struct unwind_info *ui)
48{
49 struct addr_location al;
50
51 return __report_module(&al, ip, ui);
52}
53
54static int entry(u64 ip, struct unwind_info *ui)
55
56{
57 struct unwind_entry e;
58 struct addr_location al;
59
60 if (__report_module(&al, ip, ui))
61 return -1;
62
63 e.ip = ip;
64 e.map = al.map;
65 e.sym = al.sym;
66
67 pr_debug("unwind: %s:ip = 0x%" PRIx64 " (0x%" PRIx64 ")\n",
68 al.sym ? al.sym->name : "''",
69 ip,
70 al.map ? al.map->map_ip(al.map, ip) : (u64) 0);
71
72 return ui->cb(&e, ui->arg);
73}
74
75static pid_t next_thread(Dwfl *dwfl, void *arg, void **thread_argp)
76{
77 /* We want only single thread to be processed. */
78 if (*thread_argp != NULL)
79 return 0;
80
81 *thread_argp = arg;
82 return dwfl_pid(dwfl);
83}
84
85static int access_dso_mem(struct unwind_info *ui, Dwarf_Addr addr,
86 Dwarf_Word *data)
87{
88 struct addr_location al;
89 ssize_t size;
90
91 thread__find_addr_map(ui->thread, ui->machine, PERF_RECORD_MISC_USER,
92 MAP__FUNCTION, addr, &al);
93 if (!al.map) {
94 pr_debug("unwind: no map for %lx\n", (unsigned long)addr);
95 return -1;
96 }
97
98 if (!al.map->dso)
99 return -1;
100
101 size = dso__data_read_addr(al.map->dso, al.map, ui->machine,
102 addr, (u8 *) data, sizeof(*data));
103
104 return !(size == sizeof(*data));
105}
106
107static bool memory_read(Dwfl *dwfl __maybe_unused, Dwarf_Addr addr, Dwarf_Word *result,
108 void *arg)
109{
110 struct unwind_info *ui = arg;
111 struct stack_dump *stack = &ui->sample->user_stack;
112 u64 start, end;
113 int offset;
114 int ret;
115
116 ret = perf_reg_value(&start, &ui->sample->user_regs, PERF_REG_SP);
117 if (ret)
118 return false;
119
120 end = start + stack->size;
121
122 /* Check overflow. */
123 if (addr + sizeof(Dwarf_Word) < addr)
124 return false;
125
126 if (addr < start || addr + sizeof(Dwarf_Word) > end) {
127 ret = access_dso_mem(ui, addr, result);
128 if (ret) {
129 pr_debug("unwind: access_mem 0x%" PRIx64 " not inside range"
130 " 0x%" PRIx64 "-0x%" PRIx64 "\n",
131 addr, start, end);
132 return false;
133 }
134 return true;
135 }
136
137 offset = addr - start;
138 *result = *(Dwarf_Word *)&stack->data[offset];
139 pr_debug("unwind: access_mem addr 0x%" PRIx64 ", val %lx, offset %d\n",
140 addr, (unsigned long)*result, offset);
141 return true;
142}
143
144static const Dwfl_Thread_Callbacks callbacks = {
145 .next_thread = next_thread,
146 .memory_read = memory_read,
147 .set_initial_registers = libdw__arch_set_initial_registers,
148};
149
150static int
151frame_callback(Dwfl_Frame *state, void *arg)
152{
153 struct unwind_info *ui = arg;
154 Dwarf_Addr pc;
155
156 if (!dwfl_frame_pc(state, &pc, NULL)) {
157 pr_err("%s", dwfl_errmsg(-1));
158 return DWARF_CB_ABORT;
159 }
160
161 return entry(pc, ui) || !(--ui->max_stack) ?
162 DWARF_CB_ABORT : DWARF_CB_OK;
163}
164
165int unwind__get_entries(unwind_entry_cb_t cb, void *arg,
166 struct machine *machine, struct thread *thread,
167 struct perf_sample *data,
168 int max_stack)
169{
170 struct unwind_info ui = {
171 .sample = data,
172 .thread = thread,
173 .machine = machine,
174 .cb = cb,
175 .arg = arg,
176 .max_stack = max_stack,
177 };
178 Dwarf_Word ip;
179 int err = -EINVAL;
180
181 if (!data->user_regs.regs)
182 return -EINVAL;
183
184 ui.dwfl = dwfl_begin(&offline_callbacks);
185 if (!ui.dwfl)
186 goto out;
187
188 err = perf_reg_value(&ip, &data->user_regs, PERF_REG_IP);
189 if (err)
190 goto out;
191
192 err = report_module(ip, &ui);
193 if (err)
194 goto out;
195
196 if (!dwfl_attach_state(ui.dwfl, EM_NONE, thread->tid, &callbacks, &ui))
197 goto out;
198
199 err = dwfl_getthread_frames(ui.dwfl, thread->tid, frame_callback, &ui);
200
201 if (err && !ui.max_stack)
202 err = 0;
203
204 out:
205 if (err)
206 pr_debug("unwind: failed with '%s'\n", dwfl_errmsg(-1));
207
208 dwfl_end(ui.dwfl);
209 return 0;
210}
diff --git a/tools/perf/util/unwind-libdw.h b/tools/perf/util/unwind-libdw.h
new file mode 100644
index 000000000000..417a1426f3ad
--- /dev/null
+++ b/tools/perf/util/unwind-libdw.h
@@ -0,0 +1,21 @@
1#ifndef __PERF_UNWIND_LIBDW_H
2#define __PERF_UNWIND_LIBDW_H
3
4#include <elfutils/libdwfl.h>
5#include "event.h"
6#include "thread.h"
7#include "unwind.h"
8
9bool libdw__arch_set_initial_registers(Dwfl_Thread *thread, void *arg);
10
11struct unwind_info {
12 Dwfl *dwfl;
13 struct perf_sample *sample;
14 struct machine *machine;
15 struct thread *thread;
16 unwind_entry_cb_t cb;
17 void *arg;
18 int max_stack;
19};
20
21#endif /* __PERF_UNWIND_LIBDW_H */
diff --git a/tools/perf/util/unwind.c b/tools/perf/util/unwind-libunwind.c
index 742f23bf35ff..bd5768d74f01 100644
--- a/tools/perf/util/unwind.c
+++ b/tools/perf/util/unwind-libunwind.c
@@ -86,7 +86,6 @@ struct unwind_info {
86 struct perf_sample *sample; 86 struct perf_sample *sample;
87 struct machine *machine; 87 struct machine *machine;
88 struct thread *thread; 88 struct thread *thread;
89 u64 sample_uregs;
90}; 89};
91 90
92#define dw_read(ptr, type, end) ({ \ 91#define dw_read(ptr, type, end) ({ \
@@ -391,30 +390,13 @@ static int access_dso_mem(struct unwind_info *ui, unw_word_t addr,
391 return !(size == sizeof(*data)); 390 return !(size == sizeof(*data));
392} 391}
393 392
394static int reg_value(unw_word_t *valp, struct regs_dump *regs, int id,
395 u64 sample_regs)
396{
397 int i, idx = 0;
398
399 if (!(sample_regs & (1 << id)))
400 return -EINVAL;
401
402 for (i = 0; i < id; i++) {
403 if (sample_regs & (1 << i))
404 idx++;
405 }
406
407 *valp = regs->regs[idx];
408 return 0;
409}
410
411static int access_mem(unw_addr_space_t __maybe_unused as, 393static int access_mem(unw_addr_space_t __maybe_unused as,
412 unw_word_t addr, unw_word_t *valp, 394 unw_word_t addr, unw_word_t *valp,
413 int __write, void *arg) 395 int __write, void *arg)
414{ 396{
415 struct unwind_info *ui = arg; 397 struct unwind_info *ui = arg;
416 struct stack_dump *stack = &ui->sample->user_stack; 398 struct stack_dump *stack = &ui->sample->user_stack;
417 unw_word_t start, end; 399 u64 start, end;
418 int offset; 400 int offset;
419 int ret; 401 int ret;
420 402
@@ -424,8 +406,7 @@ static int access_mem(unw_addr_space_t __maybe_unused as,
424 return 0; 406 return 0;
425 } 407 }
426 408
427 ret = reg_value(&start, &ui->sample->user_regs, PERF_REG_SP, 409 ret = perf_reg_value(&start, &ui->sample->user_regs, PERF_REG_SP);
428 ui->sample_uregs);
429 if (ret) 410 if (ret)
430 return ret; 411 return ret;
431 412
@@ -438,8 +419,9 @@ static int access_mem(unw_addr_space_t __maybe_unused as,
438 if (addr < start || addr + sizeof(unw_word_t) >= end) { 419 if (addr < start || addr + sizeof(unw_word_t) >= end) {
439 ret = access_dso_mem(ui, addr, valp); 420 ret = access_dso_mem(ui, addr, valp);
440 if (ret) { 421 if (ret) {
441 pr_debug("unwind: access_mem %p not inside range %p-%p\n", 422 pr_debug("unwind: access_mem %p not inside range"
442 (void *)addr, (void *)start, (void *)end); 423 " 0x%" PRIx64 "-0x%" PRIx64 "\n",
424 (void *) addr, start, end);
443 *valp = 0; 425 *valp = 0;
444 return ret; 426 return ret;
445 } 427 }
@@ -448,8 +430,8 @@ static int access_mem(unw_addr_space_t __maybe_unused as,
448 430
449 offset = addr - start; 431 offset = addr - start;
450 *valp = *(unw_word_t *)&stack->data[offset]; 432 *valp = *(unw_word_t *)&stack->data[offset];
451 pr_debug("unwind: access_mem addr %p, val %lx, offset %d\n", 433 pr_debug("unwind: access_mem addr %p val %lx, offset %d\n",
452 (void *)addr, (unsigned long)*valp, offset); 434 (void *) addr, (unsigned long)*valp, offset);
453 return 0; 435 return 0;
454} 436}
455 437
@@ -459,6 +441,7 @@ static int access_reg(unw_addr_space_t __maybe_unused as,
459{ 441{
460 struct unwind_info *ui = arg; 442 struct unwind_info *ui = arg;
461 int id, ret; 443 int id, ret;
444 u64 val;
462 445
463 /* Don't support write, I suspect we don't need it. */ 446 /* Don't support write, I suspect we don't need it. */
464 if (__write) { 447 if (__write) {
@@ -471,16 +454,17 @@ static int access_reg(unw_addr_space_t __maybe_unused as,
471 return 0; 454 return 0;
472 } 455 }
473 456
474 id = unwind__arch_reg_id(regnum); 457 id = libunwind__arch_reg_id(regnum);
475 if (id < 0) 458 if (id < 0)
476 return -EINVAL; 459 return -EINVAL;
477 460
478 ret = reg_value(valp, &ui->sample->user_regs, id, ui->sample_uregs); 461 ret = perf_reg_value(&val, &ui->sample->user_regs, id);
479 if (ret) { 462 if (ret) {
480 pr_err("unwind: can't read reg %d\n", regnum); 463 pr_err("unwind: can't read reg %d\n", regnum);
481 return ret; 464 return ret;
482 } 465 }
483 466
467 *valp = (unw_word_t) val;
484 pr_debug("unwind: reg %d, val %lx\n", regnum, (unsigned long)*valp); 468 pr_debug("unwind: reg %d, val %lx\n", regnum, (unsigned long)*valp);
485 return 0; 469 return 0;
486} 470}
@@ -563,7 +547,7 @@ static int get_entries(struct unwind_info *ui, unwind_entry_cb_t cb,
563 unw_word_t ip; 547 unw_word_t ip;
564 548
565 unw_get_reg(&c, UNW_REG_IP, &ip); 549 unw_get_reg(&c, UNW_REG_IP, &ip);
566 ret = entry(ip, ui->thread, ui->machine, cb, arg); 550 ret = ip ? entry(ip, ui->thread, ui->machine, cb, arg) : 0;
567 } 551 }
568 552
569 unw_destroy_addr_space(addr_space); 553 unw_destroy_addr_space(addr_space);
@@ -572,13 +556,11 @@ static int get_entries(struct unwind_info *ui, unwind_entry_cb_t cb,
572 556
573int unwind__get_entries(unwind_entry_cb_t cb, void *arg, 557int unwind__get_entries(unwind_entry_cb_t cb, void *arg,
574 struct machine *machine, struct thread *thread, 558 struct machine *machine, struct thread *thread,
575 u64 sample_uregs, struct perf_sample *data, 559 struct perf_sample *data, int max_stack)
576 int max_stack)
577{ 560{
578 unw_word_t ip; 561 u64 ip;
579 struct unwind_info ui = { 562 struct unwind_info ui = {
580 .sample = data, 563 .sample = data,
581 .sample_uregs = sample_uregs,
582 .thread = thread, 564 .thread = thread,
583 .machine = machine, 565 .machine = machine,
584 }; 566 };
@@ -587,7 +569,7 @@ int unwind__get_entries(unwind_entry_cb_t cb, void *arg,
587 if (!data->user_regs.regs) 569 if (!data->user_regs.regs)
588 return -EINVAL; 570 return -EINVAL;
589 571
590 ret = reg_value(&ip, &data->user_regs, PERF_REG_IP, sample_uregs); 572 ret = perf_reg_value(&ip, &data->user_regs, PERF_REG_IP);
591 if (ret) 573 if (ret)
592 return ret; 574 return ret;
593 575
@@ -595,5 +577,5 @@ int unwind__get_entries(unwind_entry_cb_t cb, void *arg,
595 if (ret) 577 if (ret)
596 return -ENOMEM; 578 return -ENOMEM;
597 579
598 return get_entries(&ui, cb, arg, max_stack); 580 return --max_stack > 0 ? get_entries(&ui, cb, arg, max_stack) : 0;
599} 581}
diff --git a/tools/perf/util/unwind.h b/tools/perf/util/unwind.h
index d5966f49e22c..b031316f221a 100644
--- a/tools/perf/util/unwind.h
+++ b/tools/perf/util/unwind.h
@@ -13,24 +13,25 @@ struct unwind_entry {
13 13
14typedef int (*unwind_entry_cb_t)(struct unwind_entry *entry, void *arg); 14typedef int (*unwind_entry_cb_t)(struct unwind_entry *entry, void *arg);
15 15
16#ifdef HAVE_LIBUNWIND_SUPPORT 16#ifdef HAVE_DWARF_UNWIND_SUPPORT
17int unwind__get_entries(unwind_entry_cb_t cb, void *arg, 17int unwind__get_entries(unwind_entry_cb_t cb, void *arg,
18 struct machine *machine, 18 struct machine *machine,
19 struct thread *thread, 19 struct thread *thread,
20 u64 sample_uregs,
21 struct perf_sample *data, int max_stack); 20 struct perf_sample *data, int max_stack);
22int unwind__arch_reg_id(int regnum); 21/* libunwind specific */
22#ifdef HAVE_LIBUNWIND_SUPPORT
23int libunwind__arch_reg_id(int regnum);
24#endif
23#else 25#else
24static inline int 26static inline int
25unwind__get_entries(unwind_entry_cb_t cb __maybe_unused, 27unwind__get_entries(unwind_entry_cb_t cb __maybe_unused,
26 void *arg __maybe_unused, 28 void *arg __maybe_unused,
27 struct machine *machine __maybe_unused, 29 struct machine *machine __maybe_unused,
28 struct thread *thread __maybe_unused, 30 struct thread *thread __maybe_unused,
29 u64 sample_uregs __maybe_unused,
30 struct perf_sample *data __maybe_unused, 31 struct perf_sample *data __maybe_unused,
31 int max_stack __maybe_unused) 32 int max_stack __maybe_unused)
32{ 33{
33 return 0; 34 return 0;
34} 35}
35#endif /* HAVE_LIBUNWIND_SUPPORT */ 36#endif /* HAVE_DWARF_UNWIND_SUPPORT */
36#endif /* __UNWIND_H */ 37#endif /* __UNWIND_H */
diff --git a/tools/perf/util/util.c b/tools/perf/util/util.c
index 42ad667bb317..9f66549562bd 100644
--- a/tools/perf/util/util.c
+++ b/tools/perf/util/util.c
@@ -1,6 +1,6 @@
1#include "../perf.h" 1#include "../perf.h"
2#include "util.h" 2#include "util.h"
3#include "fs.h" 3#include <api/fs/fs.h>
4#include <sys/mman.h> 4#include <sys/mman.h>
5#ifdef HAVE_BACKTRACE_SUPPORT 5#ifdef HAVE_BACKTRACE_SUPPORT
6#include <execinfo.h> 6#include <execinfo.h>