From b4b7ff4b08e691656c9d77c758fc355833128ac0 Mon Sep 17 00:00:00 2001
From: André Fabian Silva Delgado <emulatorman@parabola.nu>
Date: Wed, 20 Jan 2016 14:01:31 -0300
Subject: Linux-libre 4.4-gnu
---
tools/perf/arch/x86/util/dwarf-regs.c | 122 ++++++++++++++++++++++++----------
tools/perf/arch/x86/util/intel-pt.c | 55 ++++++++++++---
2 files changed, 135 insertions(+), 42 deletions(-)
(limited to 'tools/perf/arch/x86/util')
diff --git a/tools/perf/arch/x86/util/dwarf-regs.c b/tools/perf/arch/x86/util/dwarf-regs.c
index a08de0a35..9223c164e 100644
--- a/tools/perf/arch/x86/util/dwarf-regs.c
+++ b/tools/perf/arch/x86/util/dwarf-regs.c
@@ -21,55 +21,109 @@
*/
#include <stddef.h>
+#include <errno.h> /* for EINVAL */
+#include <string.h> /* for strcmp */
+#include <linux/ptrace.h> /* for struct pt_regs */
+#include <linux/kernel.h> /* for offsetof */
#include <dwarf-regs.h>
/*
- * Generic dwarf analysis helpers
+ * See arch/x86/kernel/ptrace.c.
+ * Different from it:
+ *
+ * - Since struct pt_regs is defined differently for user and kernel,
+ * but we want to use 'ax, bx' instead of 'rax, rbx' (which is struct
+ * field name of user's pt_regs), we make REG_OFFSET_NAME to accept
+ * both string name and reg field name.
+ *
+ * - Since accessing x86_32's pt_regs from x86_64 building is difficult
+ * and vise versa, we simply fill offset with -1, so
+ * get_arch_regstr() still works but regs_query_register_offset()
+ * returns error.
+ * The only inconvenience caused by it now is that we are not allowed
+ * to generate BPF prologue for a x86_64 kernel if perf is built for
+ * x86_32. This is really a rare usecase.
+ *
+ * - Order is different from kernel's ptrace.c for get_arch_regstr(). Use
+ * the order defined by dwarf.
*/
-#define X86_32_MAX_REGS 8
-const char *x86_32_regs_table[X86_32_MAX_REGS] = {
- "%ax",
- "%cx",
- "%dx",
- "%bx",
- "$stack", /* Stack address instead of %sp */
- "%bp",
- "%si",
- "%di",
+struct pt_regs_offset {
+ const char *name;
+ int offset;
+};
+
+#define REG_OFFSET_END {.name = NULL, .offset = 0}
+
+#ifdef __x86_64__
+# define REG_OFFSET_NAME_64(n, r) {.name = n, .offset = offsetof(struct pt_regs, r)}
+# define REG_OFFSET_NAME_32(n, r) {.name = n, .offset = -1}
+#else
+# define REG_OFFSET_NAME_64(n, r) {.name = n, .offset = -1}
+# define REG_OFFSET_NAME_32(n, r) {.name = n, .offset = offsetof(struct pt_regs, r)}
+#endif
+
+static const struct pt_regs_offset x86_32_regoffset_table[] = {
+ REG_OFFSET_NAME_32("%ax", eax),
+ REG_OFFSET_NAME_32("%cx", ecx),
+ REG_OFFSET_NAME_32("%dx", edx),
+ REG_OFFSET_NAME_32("%bx", ebx),
+ REG_OFFSET_NAME_32("$stack", esp), /* Stack address instead of %sp */
+ REG_OFFSET_NAME_32("%bp", ebp),
+ REG_OFFSET_NAME_32("%si", esi),
+ REG_OFFSET_NAME_32("%di", edi),
+ REG_OFFSET_END,
};
-#define X86_64_MAX_REGS 16
-const char *x86_64_regs_table[X86_64_MAX_REGS] = {
- "%ax",
- "%dx",
- "%cx",
- "%bx",
- "%si",
- "%di",
- "%bp",
- "%sp",
- "%r8",
- "%r9",
- "%r10",
- "%r11",
- "%r12",
- "%r13",
- "%r14",
- "%r15",
+static const struct pt_regs_offset x86_64_regoffset_table[] = {
+ REG_OFFSET_NAME_64("%ax", rax),
+ REG_OFFSET_NAME_64("%dx", rdx),
+ REG_OFFSET_NAME_64("%cx", rcx),
+ REG_OFFSET_NAME_64("%bx", rbx),
+ REG_OFFSET_NAME_64("%si", rsi),
+ REG_OFFSET_NAME_64("%di", rdi),
+ REG_OFFSET_NAME_64("%bp", rbp),
+ REG_OFFSET_NAME_64("%sp", rsp),
+ REG_OFFSET_NAME_64("%r8", r8),
+ REG_OFFSET_NAME_64("%r9", r9),
+ REG_OFFSET_NAME_64("%r10", r10),
+ REG_OFFSET_NAME_64("%r11", r11),
+ REG_OFFSET_NAME_64("%r12", r12),
+ REG_OFFSET_NAME_64("%r13", r13),
+ REG_OFFSET_NAME_64("%r14", r14),
+ REG_OFFSET_NAME_64("%r15", r15),
+ REG_OFFSET_END,
};
/* TODO: switching by dwarf address size */
#ifdef __x86_64__
-#define ARCH_MAX_REGS X86_64_MAX_REGS
-#define arch_regs_table x86_64_regs_table
+#define regoffset_table x86_64_regoffset_table
#else
-#define ARCH_MAX_REGS X86_32_MAX_REGS
-#define arch_regs_table x86_32_regs_table
+#define regoffset_table x86_32_regoffset_table
#endif
+/* Minus 1 for the ending REG_OFFSET_END */
+#define ARCH_MAX_REGS ((sizeof(regoffset_table) / sizeof(regoffset_table[0])) - 1)
+
/* Return architecture dependent register string (for kprobe-tracer) */
const char *get_arch_regstr(unsigned int n)
{
- return (n < ARCH_MAX_REGS) ? arch_regs_table[n] : NULL;
+ return (n < ARCH_MAX_REGS) ? regoffset_table[n].name : NULL;
+}
+
+/* Reuse code from arch/x86/kernel/ptrace.c */
+/**
+ * regs_query_register_offset() - query register offset from its name
+ * @name: the name of a register
+ *
+ * regs_query_register_offset() returns the offset of a register in struct
+ * pt_regs from its name. If the name is invalid, this returns -EINVAL;
+ */
+int regs_query_register_offset(const char *name)
+{
+ const struct pt_regs_offset *roff;
+ for (roff = regoffset_table; roff->name != NULL; roff++)
+ if (!strcmp(roff->name, name))
+ return roff->offset;
+ return -EINVAL;
}
diff --git a/tools/perf/arch/x86/util/intel-pt.c b/tools/perf/arch/x86/util/intel-pt.c
index 2ca10d796..b02af064f 100644
--- a/tools/perf/arch/x86/util/intel-pt.c
+++ b/tools/perf/arch/x86/util/intel-pt.c
@@ -624,13 +624,49 @@ static int intel_pt_recording_options(struct auxtrace_record *itr,
* threads.
*/
if (have_timing_info && !cpu_map__empty(cpus)) {
- err = intel_pt_track_switches(evlist);
- if (err == -EPERM)
- pr_debug2("Unable to select sched:sched_switch\n");
- else if (err)
- return err;
- else
- ptr->have_sched_switch = 1;
+ if (perf_can_record_switch_events()) {
+ bool cpu_wide = !target__none(&opts->target) &&
+ !target__has_task(&opts->target);
+
+ if (!cpu_wide && perf_can_record_cpu_wide()) {
+ struct perf_evsel *switch_evsel;
+
+ err = parse_events(evlist, "dummy:u", NULL);
+ if (err)
+ return err;
+
+ switch_evsel = perf_evlist__last(evlist);
+
+ switch_evsel->attr.freq = 0;
+ switch_evsel->attr.sample_period = 1;
+ switch_evsel->attr.context_switch = 1;
+
+ switch_evsel->system_wide = true;
+ switch_evsel->no_aux_samples = true;
+ switch_evsel->immediate = true;
+
+ perf_evsel__set_sample_bit(switch_evsel, TID);
+ perf_evsel__set_sample_bit(switch_evsel, TIME);
+ perf_evsel__set_sample_bit(switch_evsel, CPU);
+
+ opts->record_switch_events = false;
+ ptr->have_sched_switch = 3;
+ } else {
+ opts->record_switch_events = true;
+ if (cpu_wide)
+ ptr->have_sched_switch = 3;
+ else
+ ptr->have_sched_switch = 2;
+ }
+ } else {
+ err = intel_pt_track_switches(evlist);
+ if (err == -EPERM)
+ pr_debug2("Unable to select sched:sched_switch\n");
+ else if (err)
+ return err;
+ else
+ ptr->have_sched_switch = 1;
+ }
}
if (intel_pt_evsel) {
@@ -663,8 +699,11 @@ static int intel_pt_recording_options(struct auxtrace_record *itr,
tracking_evsel->attr.sample_period = 1;
/* In per-cpu case, always need the time of mmap events etc */
- if (!cpu_map__empty(cpus))
+ if (!cpu_map__empty(cpus)) {
perf_evsel__set_sample_bit(tracking_evsel, TIME);
+ /* And the CPU for switch events */
+ perf_evsel__set_sample_bit(tracking_evsel, CPU);
+ }
}
/*
--
cgit v1.2.3-54-g00ecf