]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
perf symbol: Lazily compute idle
authorIan Rogers <irogers@google.com>
Tue, 2 Jun 2026 15:25:16 +0000 (08:25 -0700)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Wed, 3 Jun 2026 19:54:09 +0000 (16:54 -0300)
Switch from an idle boolean to a helper symbol__is_idle function. In
the function lazily compute whether a symbol is an idle function
taking into consideration the kernel version and architecture of the
machine. As symbols__insert no longer needs to know if a symbol is for
the kernel, remove the argument.

To protect against drop-filtering of legitimate setup, online, or hotplug
management functions (such as intel_idle_init), x86 matches are strictly
constrained to exact known run-loops (intel_idle, intel_idle_irq,
mwait_idle, mwait_idle_with_hints).

If the target environment OS release is unresolvable (such as on guest
traces), default to treating psw_idle as idle to prevent false
negatives and match legacy trace behavior safely.

This change is inspired by mailing list discussion, particularly from
Thomas Richter <tmricht@linux.ibm.com> and Heiko Carstens
<hca@linux.ibm.com>:
https://lore.kernel.org/lkml/20260219113850.354271-1-tmricht@linux.ibm.com/

Assisted-by: Gemini:gemini-3.1-pro-preview
Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Honglei Wang <jameshongleiwang@126.com>
Cc: Jan Polensky <japo@linux.ibm.com>
Cc: Sumanth Korikkar <sumanthk@linux.ibm.com>
Cc: Thomas Richter <tmricht@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/symbol-elf.c
tools/perf/util/symbol.c
tools/perf/util/symbol.h

index c5ed5e0519764239d2d38becef52f4687e3a6246..186e6d92ac3d774244bfadb9c14b33968957aa53 100644 (file)
@@ -1734,7 +1734,7 @@ dso__load_sym_internal(struct dso *dso, struct map *map, struct symsrc *syms_ss,
 
                arch__sym_update(f, &sym);
 
-               __symbols__insert(dso__symbols(curr_dso), f, dso__kernel(dso));
+               __symbols__insert(dso__symbols(curr_dso), f);
                nr++;
        }
        dso__put(curr_dso);
index ddd3106b03b167742dacac5a4db4a9de988679c4..0c46b24ee0986059a66e15419abfa9029bf54ee1 100644 (file)
@@ -50,7 +50,6 @@
 
 static int dso__load_kernel_sym(struct dso *dso, struct map *map);
 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map);
-static bool symbol__compute_is_idle(const char *name);
 
 int vmlinux_path__nr_entries;
 char **vmlinux_path;
@@ -379,7 +378,7 @@ void symbol__set_ifunc_alias(struct symbol *sym, bool ifunc_alias)
 
 static void symbol__set_idle(struct symbol *sym, bool idle)
 {
-       uint16_t old_flags = atomic_load(&sym->flags);
+       uint16_t old_flags = atomic_load_explicit(&sym->flags, memory_order_relaxed);
        uint16_t new_flags;
        uint16_t idle_val = idle ? SYMBOL_IDLE__IDLE : SYMBOL_IDLE__NOT_IDLE;
 
@@ -401,8 +400,7 @@ void symbols__delete(struct rb_root_cached *symbols)
        }
 }
 
-void __symbols__insert(struct rb_root_cached *symbols,
-                      struct symbol *sym, bool kernel)
+void __symbols__insert(struct rb_root_cached *symbols, struct symbol *sym)
 {
        struct rb_node **p = &symbols->rb_root.rb_node;
        struct rb_node *parent = NULL;
@@ -410,17 +408,6 @@ void __symbols__insert(struct rb_root_cached *symbols,
        struct symbol *s;
        bool leftmost = true;
 
-       if (kernel) {
-               const char *name = sym->name;
-               /*
-                * ppc64 uses function descriptors and appends a '.' to the
-                * start of every instruction address. Remove it.
-                */
-               if (name[0] == '.')
-                       name++;
-               symbol__set_idle(sym, symbol__compute_is_idle(name));
-       }
-
        while (*p != NULL) {
                parent = *p;
                s = rb_entry(parent, struct symbol, rb_node);
@@ -437,7 +424,7 @@ void __symbols__insert(struct rb_root_cached *symbols,
 
 void symbols__insert(struct rb_root_cached *symbols, struct symbol *sym)
 {
-       __symbols__insert(symbols, sym, false);
+       __symbols__insert(symbols, sym);
 }
 
 static struct symbol *symbols__find(struct rb_root_cached *symbols, u64 ip)
@@ -598,7 +585,7 @@ void dso__reset_find_symbol_cache(struct dso *dso)
 
 void dso__insert_symbol(struct dso *dso, struct symbol *sym)
 {
-       __symbols__insert(dso__symbols(dso), sym, dso__kernel(dso));
+       __symbols__insert(dso__symbols(dso), sym);
 
        /* update the symbol cache if necessary */
        if (dso__last_find_result_addr(dso) >= sym->start &&
@@ -760,57 +747,120 @@ out:
        return err;
 }
 
-bool symbol__is_idle(struct symbol *sym,
-                    const struct dso *dso __maybe_unused,
-                    struct perf_env *env __maybe_unused)
-{
-       uint16_t flags = atomic_load_explicit(&sym->flags, memory_order_relaxed);
-       uint16_t idle_val = (flags & SYMBOL_FLAG_IDLE_MASK) >> SYMBOL_FLAG_IDLE_SHIFT;
-
-       return idle_val == SYMBOL_IDLE__IDLE;
-}
 
 /*
  * These are symbols in the kernel image, so make sure that
  * sym is from a kernel DSO.
  */
-static bool symbol__compute_is_idle(const char *name)
+static int sym_name_cmp(const void *a, const void *b)
+{
+       const char *name = a;
+       const char *const *sym = b;
+
+       return strcmp(name, *sym);
+}
+
+static bool match_x86_idle_routine(const char *name, const char *base)
 {
-       const char * const idle_symbols[] = {
+       if (strstarts(name, base)) {
+               size_t len = strlen(base);
+
+               if (name[len] == '\0' || name[len] == '.')
+                       return true;
+       }
+       return false;
+}
+
+bool symbol__is_idle(struct symbol *sym, const struct dso *dso, struct perf_env *env)
+{
+       static const char * const idle_symbols[] = {
                "acpi_idle_do_entry",
                "acpi_processor_ffh_cstate_enter",
                "arch_cpu_idle",
                "cpu_idle",
                "cpu_startup_entry",
-               "idle_cpu",
-               "intel_idle",
-               "intel_idle_ibrs",
                "default_idle",
-               "native_safe_halt",
                "enter_idle",
                "exit_idle",
-               "mwait_idle",
-               "mwait_idle_with_hints",
-               "mwait_idle_with_hints.constprop.0",
+               "idle_cpu",
+               "native_safe_halt",
                "poll_idle",
-               "ppc64_runlatch_off",
                "pseries_dedicated_idle_sleep",
-               "psw_idle",
-               "psw_idle_exit",
-               NULL
        };
-       int i;
-       static struct strlist *idle_symbols_list;
+       const char *name = sym->name;
+       uint16_t e_machine;
 
-       if (idle_symbols_list)
-               return strlist__has_entry(idle_symbols_list, name);
+       {
+               uint16_t flags = atomic_load_explicit(&sym->flags, memory_order_relaxed);
+               uint16_t idle_val = (flags & SYMBOL_FLAG_IDLE_MASK) >> SYMBOL_FLAG_IDLE_SHIFT;
 
-       idle_symbols_list = strlist__new(NULL, NULL);
+               if (idle_val != SYMBOL_IDLE__UNKNOWN)
+                       return idle_val == SYMBOL_IDLE__IDLE;
+       }
 
-       for (i = 0; idle_symbols[i]; i++)
-               strlist__add(idle_symbols_list, idle_symbols[i]);
+       if (!dso || dso__kernel(dso) == DSO_SPACE__USER) {
+               symbol__set_idle(sym, /*idle=*/false);
+               return false;
+       }
+
+       /*
+        * ppc64 uses function descriptors and appends a '.' to the
+        * start of every instruction address. Remove it.
+        */
+       if (name[0] == '.')
+               name++;
+
+       if (bsearch(name, idle_symbols, ARRAY_SIZE(idle_symbols),
+                   sizeof(idle_symbols[0]), sym_name_cmp)) {
+               symbol__set_idle(sym, /*idle=*/true);
+               return true;
+       }
+
+       e_machine = (env && env->arch) ? perf_env__e_machine(env, NULL) : EM_NONE;
+       if (e_machine == EM_NONE && dso)
+               e_machine = dso__e_machine((struct dso *)dso, NULL, NULL);
+       if (e_machine == EM_NONE && env)
+               e_machine = perf_env__e_machine(env, NULL);
+
+       if (e_machine == EM_386 || e_machine == EM_X86_64) {
+               if (match_x86_idle_routine(name, "intel_idle") ||
+                   match_x86_idle_routine(name, "intel_idle_irq") ||
+                   match_x86_idle_routine(name, "intel_idle_ibrs") ||
+                   match_x86_idle_routine(name, "mwait_idle") ||
+                   match_x86_idle_routine(name, "mwait_idle_with_hints")) {
+                       symbol__set_idle(sym, /*idle=*/true);
+                       return true;
+               }
+       }
+
+       if (e_machine == EM_PPC64 && !strcmp(name, "ppc64_runlatch_off")) {
+               symbol__set_idle(sym, /*idle=*/true);
+               return true;
+       }
+
+       if (e_machine == EM_S390 && strstarts(name, "psw_idle")) {
+               int major = 0, minor = 0;
+               const char *release = env ? perf_env__os_release(env) : NULL;
 
-       return strlist__has_entry(idle_symbols_list, name);
+               /*
+                * If we can't determine the release (e.g. unpopulated guest traces),
+                * default to idle.
+                */
+               if (!release) {
+                       symbol__set_idle(sym, /*idle=*/true);
+                       return true;
+               }
+
+               /* Before v6.10, s390 used psw_idle. */
+               if (sscanf(release, "%d.%d", &major, &minor) == 2 &&
+                   (major < 6 || (major == 6 && minor < 10))) {
+                       symbol__set_idle(sym, /*idle=*/true);
+                       return true;
+               }
+       }
+
+       symbol__set_idle(sym, /*idle=*/false);
+       return false;
 }
 
 static int map__process_kallsym_symbol(void *arg, const char *name,
@@ -839,7 +889,7 @@ static int map__process_kallsym_symbol(void *arg, const char *name,
         * We will pass the symbols to the filter later, in
         * map__split_kallsyms, when we have split the maps per module
         */
-       __symbols__insert(root, sym, !strchr(name, '['));
+       __symbols__insert(root, sym);
 
        return 0;
 }
index 16a27074a474f52958d1d9c3f203fcf0c394ec39..a7152533570378cbd7ed899ecbdc5dca5d83c8c1 100644 (file)
@@ -76,7 +76,6 @@ enum symbol_idle_kind {
 #define SYMBOL_FLAG_INLINED         (1 << 11)
 #define SYMBOL_FLAG_ANNOTATE2       (1 << 12)
 #define SYMBOL_FLAG_IFUNC_ALIAS     (1 << 13)
-
 /**
  * A symtab entry. When allocated this may be preceded by an annotation (see
  * symbol__annotation) and/or a browser_index (see symbol__browser_index).
@@ -257,8 +256,7 @@ int dso__synthesize_plt_symbols(struct dso *dso, struct symsrc *ss);
 
 char *dso__demangle_sym(struct dso *dso, int kmodule, const char *elf_name);
 
-void __symbols__insert(struct rb_root_cached *symbols, struct symbol *sym,
-                      bool kernel);
+void __symbols__insert(struct rb_root_cached *symbols, struct symbol *sym);
 void symbols__insert(struct rb_root_cached *symbols, struct symbol *sym);
 void symbols__fixup_duplicate(struct rb_root_cached *symbols);
 void symbols__fixup_end(struct rb_root_cached *symbols, bool is_kallsyms);