perf probe: Attempt to improve C++ probing

index c006592..d69f679 100644
@@ -43,7 +43,7 @@
 #include "util/probe-file.h"
 
 #define DEFAULT_VAR_FILTER "!__k???tab_* & !__crc_*"
-#define DEFAULT_FUNC_FILTER "!_*"
+#define DEFAULT_FUNC_FILTER "!_* | _Z*"
 #define DEFAULT_LIST_FILTER "*"
 
 /* Session management structure */

index e1dbc98..39a2d47 100644
@@ -1237,6 +1237,59 @@ static bool is_c_func_name(const char *name)
 	return true;
 }
 
+/* Symbols in demangled CXX function names */
+static inline bool is_cxx_symbol(const char symbol)
+{
+	switch (symbol) {
+	case '_':
+	case ' ':
+	case '&':
+	case '*':
+	case '@':
+	case ',':
+	case ':':
+	case '<':
+	case '>':
+	case '(':
+	case ')':
+		return true;
+	default:
+		return false;
+	}
+}
+
+/* Is name a C++ name? */
+static bool is_cxx_func_name(const char *name)
+{
+	/* C name or a mangled name */
+	if (is_c_func_name(name))
+		return true;
+	while (*++name != '\0') {
+		if (!isalpha(*name) && !isdigit(*name) && !is_cxx_symbol(*name))
+			return false;
+	}
+	return true;
+}
+
+/*
+ * Find the first ':' that isn't part of a C++ namespace or class
+ * name.
+ */
+static char *first_non_cxx_ns(char *name)
+{
+	while (*name) {
+		char cur = *name, nxt = *(name + 1);
+
+		if (cur == ':' && nxt == ':')
+			name += 2;
+		else if (cur == ':')
+			return name;
+
+		name += 1;
+	}
+	return NULL;
+}
+
 /*
  * Stuff 'lr' according to the line range described by 'arg'.
  * The line range syntax is described by:

@@ -1255,7 +1308,7 @@ int parse_line_range_desc(const char *arg, struct line_range *lr)
 	lr->start = 0;
 	lr->end = INT_MAX;
 
-	range = strchr(name, ':');
+	range = first_non_cxx_ns(name);
 	if (range) {
 		*range++ = '\0';
 

@@ -1309,6 +1362,8 @@ int parse_line_range_desc(const char *arg, struct line_range *lr)
 		lr->file = name;
 	else if (is_c_func_name(name))/* We reuse it for checking funcname */
 		lr->function = name;
+	else if (is_cxx_func_name(name))
+		lr->function = name;
 	else {	/* Invalid name */
 		semantic_error("'%s' is not a valid function name.\n", name);
 		err = -EINVAL;

index c37fbef..c73dccc 100644
@@ -96,7 +96,7 @@ static int debuginfo__init_offline_dwarf(struct debuginfo *dbg,
 	return -ENOENT;
 }
 
-static struct debuginfo *__debuginfo__new(const char *path)
+static struct debuginfo *__debuginfo__new(const char *path, struct dso *dso)
 {
 	struct debuginfo *dbg = zalloc(sizeof(*dbg));
 	if (!dbg)

@@ -104,8 +104,10 @@ static struct debuginfo *__debuginfo__new(const char *path)
 
 	if (debuginfo__init_offline_dwarf(dbg, path) < 0)
 		zfree(&dbg);
-	if (dbg)
+	if (dbg) {
 		pr_debug("Open Debuginfo file: %s\n", path);
+		dbg->dso = dso__get(dso);
+	}
 	return dbg;
 }
 

@@ -135,13 +137,15 @@ struct debuginfo *debuginfo__new(const char *path)
 		if (dso__read_binary_type_filename(dso, *type, &nil,
 						   buf, PATH_MAX) < 0)
 			continue;
-		dinfo = __debuginfo__new(buf);
+		dinfo = __debuginfo__new(buf, dso);
 	}
-	dso__put(dso);
 
 out:
 	/* if failed to open all distro debuginfo, open given binary */
-	return dinfo ? : __debuginfo__new(path);
+	if (!dinfo)
+		dinfo = __debuginfo__new(path, dso);
+	dso__put(dso);
+	return dinfo;
 }
 
 void debuginfo__delete(struct debuginfo *dbg)

@@ -149,6 +153,7 @@ void debuginfo__delete(struct debuginfo *dbg)
 	if (dbg) {
 		if (dbg->dwfl)
 			dwfl_end(dbg->dwfl);
+		dso__put(dbg->dso);
 		free(dbg);
 	}
 }

@@ -167,6 +172,32 @@ static struct probe_trace_arg_ref *alloc_trace_arg_ref(long offs)
 }
 
 /*
+ * Check if the the demangled linkage_name matches the function. E.g. the
+ * linkage name of _ZNKSt6vectorIiSaIiEE4sizeEv matching the c++ function name
+ * of std::vector<int, std::allocator<int> >::size() const.
+ */
+static bool matches_demangled(struct debuginfo *dbg, Dwarf_Die *dw_die,
+			      const char *function)
+{
+	const char *name;
+	char *demangled;
+	bool res;
+
+	name = die_get_linkage_name(dw_die);
+	if (!name)
+		return false;
+
+	demangled = dso__demangle_sym(dbg->dso, 0, name);
+	if (!demangled)
+		return false;
+
+	res = strglobmatch(demangled, function);
+	free(demangled);
+	return res;
+}
+
+
+/*
  * Convert a location into trace_arg.
  * If tvar == NULL, this just checks variable can be converted.
  * If fentry == true and vr_die is a parameter, do huristic search

@@ -975,6 +1006,7 @@ static int probe_point_inline_cb(Dwarf_Die *in_die, void *data)
 struct dwarf_callback_param {
 	void *data;
 	int retval;
+	struct debuginfo *dbg;
 };
 
 /* Search function from function name */

@@ -1721,7 +1753,8 @@ static int line_range_search_cb(Dwarf_Die *sp_die, void *data)
 		return DWARF_CB_OK;
 
 	if (die_is_func_def(sp_die) &&
-	    die_match_name(sp_die, lr->function)) {
+	    (die_match_name(sp_die, lr->function) ||
+	     matches_demangled(param->dbg, sp_die, lr->function))) {
 		lf->fname = dwarf_decl_file(sp_die);
 		dwarf_decl_line(sp_die, &lr->offset);
 		pr_debug("fname: %s, lineno:%d\n", lf->fname, lr->offset);

@@ -1744,9 +1777,11 @@ static int line_range_search_cb(Dwarf_Die *sp_die, void *data)
 	return DWARF_CB_OK;
 }
 
-static int find_line_range_by_func(struct line_finder *lf)
+static int find_line_range_by_func(struct debuginfo *dbg,
+				   struct line_finder *lf)
 {
-	struct dwarf_callback_param param = {.data = (void *)lf, .retval = 0};
+	struct dwarf_callback_param param = {
+		.data = (void *)lf, .retval = 0, .dbg = dbg};
 	dwarf_getfuncs(&lf->cu_die, line_range_search_cb, &param, 0);
 	return param.retval;
 }

@@ -1766,7 +1801,7 @@ int debuginfo__find_line_range(struct debuginfo *dbg, struct line_range *lr)
 			.function = lr->function, .file = lr->file,
 			.cu_die = &lf.cu_die, .sp_die = &lf.sp_die, .found = 0};
 		struct dwarf_callback_param line_range_param = {
-			.data = (void *)&lf, .retval = 0};
+			.data = (void *)&lf, .retval = 0, .dbg = dbg};
 
 		dwarf_getpubnames(dbg->dbg, pubname_search_cb,
 				  &pubname_param, 0);

@@ -1796,7 +1831,7 @@ int debuginfo__find_line_range(struct debuginfo *dbg, struct line_range *lr)
 
 		if (!lr->file || lf.fname) {
 			if (lr->function)
-				ret = find_line_range_by_func(&lf);
+				ret = find_line_range_by_func(dbg, &lf);
 			else {
 				lf.lno_s = lr->start;
 				lf.lno_e = lr->end;

index 1625298..e28acbd 100644
@@ -14,6 +14,8 @@
 #define PROBE_ARG_VARS		"$vars"
 #define PROBE_ARG_PARAMS	"$params"
 
+struct dso;
+
 static inline int is_c_varname(const char *name)
 {
 	/* TODO */

@@ -32,6 +34,7 @@ struct debuginfo {
 	Dwfl_Module	*mod;
 	Dwfl		*dwfl;
 	Dwarf_Addr	bias;
+	struct dso	*dso;
 };
 
 /* This also tries to open distro debuginfo */

index 39a2d47..97d6b6a 100644
@@ -1407,6 +1407,22 @@ static int parse_perf_probe_event_name(char **arg, struct perf_probe_event *pev)
 	return 0;
 }
 
+/* Split the function name from  @file, :line, %return but be C++ aware */
+static char *split_func_name(char *arg)
+{
+	char *ptr = arg;
+
+	while ((ptr = strpbrk_esc(ptr, ";:+@%"))) {
+		if (ptr[0] == ':' && ptr[1] == ':') {
+			ptr += 2;
+			continue;
+		}
+		return ptr;
+	}
+
+	return NULL;
+}
+
 /* Parse probepoint definition. */
 static int parse_perf_probe_point(char *arg, struct perf_probe_event *pev)
 {

@@ -1486,7 +1502,7 @@ static int parse_perf_probe_point(char *arg, struct perf_probe_event *pev)
 			file_spec = true;
 	}
 
-	ptr = strpbrk_esc(arg, ";:+@%");
+	ptr = split_func_name(arg);
 	if (ptr) {
 		nc = *ptr;
 		*ptr++ = '\0';

@@ -1726,7 +1742,7 @@ int parse_perf_probe_command(const char *cmd, struct perf_probe_event *pev)
 	char **argv;
 	int argc, i, ret = 0;
 
-	argv = argv_split(cmd, &argc);
+	argv = argv_split_cxx(cmd, &argc);
 	if (!argv) {
 		pr_debug("Failed to split arguments.\n");
 		return -ENOMEM;

69 lines not on this page · raw message

index 4c68a09..d32de6f 100644
@@ -8,6 +8,7 @@
 
 s64 perf_atoll(const char *str);
 char **argv_split(const char *str, int *argcp);
+char **argv_split_cxx(const char *str, int *argvcp);
 void argv_free(char **argv);
 bool strglobmatch(const char *str, const char *pat);
 bool strglobmatch_nocase(const char *str, const char *pat);

index c73dccc..4ba4b18 100644
@@ -760,6 +760,7 @@ struct find_scope_param {
 	int line;
 	int diff;
 	Dwarf_Die *die_mem;
+	struct debuginfo *dbg;
 	bool found;
 };
 

@@ -777,7 +778,8 @@ static int find_best_scope_cb(Dwarf_Die *fn_die, void *data)
 	}
 	/* If the function name is given, that's what user expects */
 	if (fsp->function) {
-		if (die_match_name(fn_die, fsp->function)) {
+		if (die_match_name(fn_die, fsp->function) ||
+			matches_demangled(fsp->dbg, fn_die, fsp->function)) {
 			memcpy(fsp->die_mem, fn_die, sizeof(Dwarf_Die));
 			fsp->found = true;
 			return 1;

@@ -795,8 +797,16 @@ static int find_best_scope_cb(Dwarf_Die *fn_die, void *data)
 	return 0;
 }
 
+/* Callback parameter with return value for libdw */
+struct dwarf_callback_param {
+	void *data;
+	int retval;
+	struct debuginfo *dbg;
+};
+
 /* Find an appropriate scope fits to given conditions */
-static Dwarf_Die *find_best_scope(struct probe_finder *pf, Dwarf_Die *die_mem)
+static Dwarf_Die *find_best_scope(struct debuginfo *dbg,
+				  struct probe_finder *pf, Dwarf_Die *die_mem)
 {
 	struct find_scope_param fsp = {
 		.function = pf->pev->point.function,

@@ -804,6 +814,7 @@ static Dwarf_Die *find_best_scope(struct probe_finder *pf, Dwarf_Die *die_mem)
 		.line = pf->lno,
 		.diff = INT_MAX,
 		.die_mem = die_mem,
+		.dbg = dbg,
 		.found = false,
 	};
 

@@ -815,7 +826,8 @@ static Dwarf_Die *find_best_scope(struct probe_finder *pf, Dwarf_Die *die_mem)
 static int probe_point_line_walker(const char *fname, int lineno,
 				   Dwarf_Addr addr, void *data)
 {
-	struct probe_finder *pf = data;
+	struct dwarf_callback_param *param = data;
+	struct probe_finder *pf = param->data;
 	Dwarf_Die *sc_die, die_mem;
 	int ret;
 

@@ -823,7 +835,7 @@ static int probe_point_line_walker(const char *fname, int lineno,
 		return 0;
 
 	pf->addr = addr;
-	sc_die = find_best_scope(pf, &die_mem);
+	sc_die = find_best_scope(param->dbg, pf, &die_mem);
 	if (!sc_die) {
 		pr_warning("Failed to find scope of probe point.\n");
 		return -ENOENT;

@@ -836,9 +848,12 @@ static int probe_point_line_walker(const char *fname, int lineno,
 }
 
 /* Find probe point from its line number */
-static int find_probe_point_by_line(struct probe_finder *pf)
+static int find_probe_point_by_line(struct debuginfo *dbg,
+				    struct probe_finder *pf)
 {
-	return die_walk_lines(&pf->cu_die, probe_point_line_walker, pf);
+	struct dwarf_callback_param param = {
+		.data = (void *)pf, .dbg = dbg, .retval = 0};
+	return die_walk_lines(&pf->cu_die, probe_point_line_walker, &param);
 }
 
 /* Find lines which match lazy pattern */

@@ -884,7 +899,8 @@ static int find_lazy_match_lines(struct intlist *list,
 static int probe_point_lazy_walker(const char *fname, int lineno,
 				   Dwarf_Addr addr, void *data)
 {
-	struct probe_finder *pf = data;
+	struct dwarf_callback_param *param = data;
+	struct probe_finder *pf = param->data;
 	Dwarf_Die *sc_die, die_mem;
 	int ret;
 

@@ -896,7 +912,7 @@ static int probe_point_lazy_walker(const char *fname, int lineno,
 		 lineno, (unsigned long long)addr);
 	pf->addr = addr;
 	pf->lno = lineno;
-	sc_die = find_best_scope(pf, &die_mem);
+	sc_die = find_best_scope(param->dbg, pf, &die_mem);
 	if (!sc_die) {
 		pr_warning("Failed to find scope of probe point.\n");
 		return -ENOENT;

@@ -912,8 +928,10 @@ static int probe_point_lazy_walker(const char *fname, int lineno,
 }
 
 /* Find probe points from lazy pattern  */
-static int find_probe_point_lazy(Dwarf_Die *sp_die, struct probe_finder *pf)
+static int find_probe_point_lazy(Dwarf_Die *sp_die,
+				 struct dwarf_callback_param *param)
 {
+	struct probe_finder *pf = param->data;
 	int ret = 0;
 	char *fpath;
 

@@ -935,7 +953,7 @@ static int find_probe_point_lazy(Dwarf_Die *sp_die, struct probe_finder *pf)
 			return ret;
 	}
 
-	return die_walk_lines(sp_die, probe_point_lazy_walker, pf);
+	return die_walk_lines(sp_die, probe_point_lazy_walker, param);
 }
 
 static void skip_prologue(Dwarf_Die *sp_die, struct probe_finder *pf)

@@ -972,13 +990,14 @@ static void skip_prologue(Dwarf_Die *sp_die, struct probe_finder *pf)
 
 static int probe_point_inline_cb(Dwarf_Die *in_die, void *data)
 {
-	struct probe_finder *pf = data;
+	struct dwarf_callback_param *param = data;
+	struct probe_finder *pf = param->data;
 	struct perf_probe_point *pp = &pf->pev->point;
 	Dwarf_Addr addr;
 	int ret;
 
 	if (pp->lazy_line)
-		ret = find_probe_point_lazy(in_die, pf);
+		ret = find_probe_point_lazy(in_die, param);
 	else {
 		/* Get probe address */
 		if (dwarf_entrypc(in_die, &addr) != 0) {

@@ -1002,13 +1021,6 @@ static int probe_point_inline_cb(Dwarf_Die *in_die, void *data)
 	return ret;
 }
 
-/* Callback parameter with return value for libdw */
-struct dwarf_callback_param {
-	void *data;
-	int retval;
-	struct debuginfo *dbg;
-};
-
 /* Search function from function name */
 static int probe_point_search_cb(Dwarf_Die *sp_die, void *data)
 {

@@ -1018,7 +1030,8 @@ static int probe_point_search_cb(Dwarf_Die *sp_die, void *data)
 
 	/* Check tag and diename */
 	if (!die_is_func_def(sp_die) ||
-	    !die_match_name(sp_die, pp->function))
+	    (!die_match_name(sp_die, pp->function) &&
+	     !matches_demangled(param->dbg, sp_die, pp->function)))
 		return DWARF_CB_OK;
 
 	/* Check declared file */

@@ -1031,7 +1044,7 @@ static int probe_point_search_cb(Dwarf_Die *sp_die, void *data)
 	if (pp->line) { /* Function relative line */
 		dwarf_decl_line(sp_die, &pf->lno);
 		pf->lno += pp->line;
-		param->retval = find_probe_point_by_line(pf);
+		param->retval = find_probe_point_by_line(param->dbg, pf);
 	} else if (die_is_func_instance(sp_die)) {
 		/* Instances always have the entry address */
 		dwarf_entrypc(sp_die, &pf->addr);

@@ -1042,7 +1055,7 @@ static int probe_point_search_cb(Dwarf_Die *sp_die, void *data)
 			param->retval = 0;
 		/* Real function */
 		} else if (pp->lazy_line)
-			param->retval = find_probe_point_lazy(sp_die, pf);
+			param->retval = find_probe_point_lazy(sp_die, param);
 		else {
 			skip_prologue(sp_die, pf);
 			pf->addr += pp->offset;

@@ -1052,7 +1065,7 @@ static int probe_point_search_cb(Dwarf_Die *sp_die, void *data)
 	} else if (!probe_conf.no_inlines) {
 		/* Inlined function: search instances */
 		param->retval = die_walk_instances(sp_die,
-					probe_point_inline_cb, (void *)pf);
+					probe_point_inline_cb, data);
 		/* This could be a non-existed inline definition */
 		if (param->retval == -ENOENT)
 			param->retval = 0;

@@ -1067,9 +1080,10 @@ static int probe_point_search_cb(Dwarf_Die *sp_die, void *data)
 	return DWARF_CB_ABORT; /* Exit; no same symbol in this CU. */
 }
 
-static int find_probe_point_by_func(struct probe_finder *pf)
+static int find_probe_point_by_func(struct debuginfo *dbg,
+				    struct probe_finder *pf)
 {
-	struct dwarf_callback_param _param = {.data = (void *)pf,
+	struct dwarf_callback_param _param = {.data = (void *)pf, .dbg = dbg,
 					      .retval = 0};
 	dwarf_getfuncs(&pf->cu_die, probe_point_search_cb, &_param, 0);
 	return _param.retval;

@@ -1080,6 +1094,7 @@ struct pubname_callback_param {
 	char *file;
 	Dwarf_Die *cu_die;
 	Dwarf_Die *sp_die;
+	struct debuginfo *dbg;
 	int found;
 };
 

@@ -1091,7 +1106,9 @@ static int pubname_search_cb(Dwarf *dbg, Dwarf_Global *gl, void *data)
 		if (dwarf_tag(param->sp_die) != DW_TAG_subprogram)
 			return DWARF_CB_OK;
 
-		if (die_match_name(param->sp_die, param->function)) {
+		if (die_match_name(param->sp_die, param->function) ||
+			matches_demangled(param->dbg, param->sp_die,
+					  param->function)) {
 			if (!dwarf_offdie(dbg, gl->cu_offset, param->cu_die))
 				return DWARF_CB_OK;
 

@@ -1128,10 +1145,12 @@ static int debuginfo__find_probe_location(struct debuginfo *dbg,
 			.file	  = pp->file,
 			.cu_die	  = &pf->cu_die,
 			.sp_die	  = &pf->sp_die,
+			.dbg      = dbg,
 			.found	  = 0,
 		};
 		struct dwarf_callback_param probe_param = {
 			.data = pf,
+			.dbg  = dbg,
 		};
 
 		dwarf_getpubnames(dbg->dbg, pubname_search_cb,

@@ -1158,12 +1177,17 @@ static int debuginfo__find_probe_location(struct debuginfo *dbg,
 
 		if (!pp->file || pf->fname) {
 			if (pp->function)
-				ret = find_probe_point_by_func(pf);
-			else if (pp->lazy_line)
-				ret = find_probe_point_lazy(&pf->cu_die, pf);
-			else {
+				ret = find_probe_point_by_func(dbg, pf);
+			else if (pp->lazy_line) {
+				struct dwarf_callback_param probe_param = {
+					.data = pf,
+					.dbg  = dbg,
+				};
+				ret = find_probe_point_lazy(&pf->cu_die,
+							    &probe_param);
+			} else {
 				pf->lno = pp->line;
-				ret = find_probe_point_by_line(pf);
+				ret = find_probe_point_by_line(dbg, pf);
 			}
 			if (ret < 0)
 				break;

@@ -1798,7 +1822,7 @@ int debuginfo__find_line_range(struct debuginfo *dbg, struct line_range *lr)
 	/* Fastpath: lookup by function name from .debug_pubnames section */
 	if (lr->function) {
 		struct pubname_callback_param pubname_param = {
-			.function = lr->function, .file = lr->file,
+			.function = lr->function, .file = lr->file, .dbg = dbg,
 			.cu_die = &lf.cu_die, .sp_die = &lf.sp_die, .found = 0};
 		struct dwarf_callback_param line_range_param = {
 			.data = (void *)&lf, .retval = 0, .dbg = dbg};

index 4ba4b18..4cfa3de 100644
@@ -1317,6 +1317,7 @@ static int add_probe_trace_event(Dwarf_Die *sc_die, struct probe_finder *pf)
 	struct perf_probe_point *pp = &pf->pev->point;
 	struct probe_trace_event *tev;
 	struct perf_probe_arg *args = NULL;
+	const char *linkage_name;
 	int ret, i;
 
 	/* Check number of tevs */

@@ -1333,6 +1334,16 @@ static int add_probe_trace_event(Dwarf_Die *sc_die, struct probe_finder *pf)
 	if (ret < 0)
 		goto end;
 
+	/*
+	 * Adding a C++ name like std::vector<int, std::allocator<int> >::at
+	 * will fail. Check if we want to use the linkage name instead.
+	 */
+	linkage_name = die_get_linkage_name(&pf->sp_die);
+	if (linkage_name && strncmp(linkage_name, "_Z", 2) == 0) {
+		free(pp->function);
+		pp->function = strdup(linkage_name);
+	}
+
 	tev->point.realname = strdup(dwarf_diename(sc_die));
 	if (!tev->point.realname) {
 		ret = -ENOMEM;