gendwarfksyms: Add support for symbol type pointers

The compiler may choose not to emit type information in DWARF for
external symbols. Clang, for example, does this for symbols not
defined in the current TU.

To provide a way to work around this issue, add support for
__gendwarfksyms_ptr_<symbol> pointers that force the compiler to emit
the necessary type information in DWARF also for the missing symbols.

Example usage:

#define GENDWARFKSYMS_PTR(sym) \
static typeof(sym) *__gendwarfksyms_ptr_##sym __used \
__section(".discard.gendwarfksyms") = &sym;

extern int external_symbol(void);
GENDWARFKSYMS_PTR(external_symbol);

Signed-off-by: Sami Tolvanen <[e-mail]>
Reviewed-by: Petr Pavlu <[e-mail]>
Signed-off-by: Masahiro Yamada <[e-mail]>

scripts/gendwarfksyms/dwarf.c

···
 	process_symbol(state, die, __process_variable);
 }
 
+static void save_symbol_ptr(struct state *state)
+{
+	Dwarf_Die ptr_type;
+	Dwarf_Die type;
+
+	if (!get_ref_die_attr(&state->die, DW_AT_type, &ptr_type) ||
+	    dwarf_tag(&ptr_type) != DW_TAG_pointer_type)
+		error("%s must be a pointer type!",
+		      get_symbol_name(&state->die));
+
+	if (!get_ref_die_attr(&ptr_type, DW_AT_type, &type))
+		error("%s pointer missing a type attribute?",
+		      get_symbol_name(&state->die));
+
+	/*
+	 * Save the symbol pointer DIE in case the actual symbol is
+	 * missing from the DWARF. Clang, for example, intentionally
+	 * omits external symbols from the debugging information.
+	 */
+	if (dwarf_tag(&type) == DW_TAG_subroutine_type)
+		symbol_set_ptr(state->sym, &type);
+	else
+		symbol_set_ptr(state->sym, &ptr_type);
+}
+
 static int process_exported_symbols(struct state *unused, struct die *cache,
 				    Dwarf_Die *die)
 {
···
 
 		state_init(&state);
 
-		if (tag == DW_TAG_subprogram)
+		if (is_symbol_ptr(get_symbol_name(&state.die)))
+			save_symbol_ptr(&state);
+		else if (tag == DW_TAG_subprogram)
 			process_subprogram(&state, &state.die);
 		else
 			process_variable(&state, &state.die);
···
 	}
 }
 
+static void process_symbol_ptr(struct symbol *sym, void *arg)
+{
+	struct state state;
+	Dwarf *dwarf = arg;
+
+	if (sym->state != SYMBOL_UNPROCESSED || !sym->ptr_die_addr)
+		return;
+
+	debug("%s", sym->name);
+	state_init(&state);
+	state.sym = sym;
+
+	if (!dwarf_die_addr_die(dwarf, (void *)sym->ptr_die_addr, &state.die))
+		error("dwarf_die_addr_die failed for symbol ptr: '%s'",
+		      sym->name);
+
+	if (dwarf_tag(&state.die) == DW_TAG_subroutine_type)
+		process_subprogram(&state, &state.die);
+	else
+		process_variable(&state, &state.die);
+
+	cache_free(&state.expansion_cache);
+}
+
 void process_cu(Dwarf_Die *cudie)
 {
 	check(process_die_container(NULL, NULL, cudie, process_exported_symbols,
 				    match_all));
+
+	symbol_for_each(process_symbol_ptr, dwarf_cu_getdwarf(cudie->cu));
 
 	cache_free(&srcfile_cache);
 }

scripts/gendwarfksyms/examples/symbolptr.c

···
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2024 Google LLC
+ *
+ * Example for symbol pointers. When compiled with Clang, gendwarfkyms
+ * uses a symbol pointer for `f`.
+ *
+ * $ clang -g -c examples/symbolptr.c -o examples/symbolptr.o
+ * $ echo -e "f\ng\np" | ./gendwarfksyms -d examples/symbolptr.o
+ */
+
+/* Kernel macros for userspace testing. */
+#ifndef __used
+#define __used __attribute__((__used__))
+#endif
+#ifndef __section
+#define __section(section) __attribute__((__section__(section)))
+#endif
+
+#define __GENDWARFKSYMS_EXPORT(sym)				\
+	static typeof(sym) *__gendwarfksyms_ptr_##sym __used	\
+		__section(".discard.gendwarfksyms") = &sym;
+
+extern void f(unsigned int arg);
+void g(int *arg);
+void g(int *arg) {}
+
+struct s;
+extern struct s *p;
+
+__GENDWARFKSYMS_EXPORT(f);
+__GENDWARFKSYMS_EXPORT(g);
+__GENDWARFKSYMS_EXPORT(p);

scripts/gendwarfksyms/gendwarfksyms.h

···
  * symbols.c
  */
 
+/* See symbols.c:is_symbol_ptr */
+#define SYMBOL_PTR_PREFIX "__gendwarfksyms_ptr_"
+#define SYMBOL_PTR_PREFIX_LEN (sizeof(SYMBOL_PTR_PREFIX) - 1)
+
 static inline unsigned int addr_hash(uintptr_t addr)
 {
 	return hash_ptr((const void *)addr);
···
 	struct hlist_node name_hash;
 	enum symbol_state state;
 	uintptr_t die_addr;
+	uintptr_t ptr_die_addr;
 	unsigned long crc;
 };
 
 typedef void (*symbol_callback_t)(struct symbol *, void *arg);
 
+bool is_symbol_ptr(const char *name);
 void symbol_read_exports(FILE *file);
 void symbol_read_symtab(int fd);
 struct symbol *symbol_get(const char *name);
+void symbol_set_ptr(struct symbol *sym, Dwarf_Die *ptr);
 void symbol_set_die(struct symbol *sym, Dwarf_Die *die);
 void symbol_set_crc(struct symbol *sym, unsigned long crc);
 void symbol_for_each(symbol_callback_t func, void *arg);

scripts/gendwarfksyms/symbols.c

···
 	return processed;
 }
 
+/*
+ * For symbols without debugging information (e.g. symbols defined in other
+ * TUs), we also match __gendwarfksyms_ptr_<symbol_name> symbols, which the
+ * kernel uses to ensure type information is present in the TU that exports
+ * the symbol. A __gendwarfksyms_ptr pointer must have the same type as the
+ * exported symbol, e.g.:
+ *
+ *   typeof(symname) *__gendwarf_ptr_symname = &symname;
+ */
+bool is_symbol_ptr(const char *name)
+{
+	return name && !strncmp(name, SYMBOL_PTR_PREFIX, SYMBOL_PTR_PREFIX_LEN);
+}
+
 static unsigned int for_each(const char *name, symbol_callback_t func,
 			     void *data)
 {
···
 
 	if (!name || !*name)
 		return 0;
+	if (is_symbol_ptr(name))
+		name += SYMBOL_PTR_PREFIX_LEN;
 
 	hash_for_each_possible_safe(symbol_names, match, tmp, name_hash,
 				    hash_str(name)) {
···
 void symbol_set_crc(struct symbol *sym, unsigned long crc)
 {
 	if (for_each(sym->name, set_crc, &crc) == 0)
+		error("no matching symbols: '%s'", sym->name);
+}
+
+static void set_ptr(struct symbol *sym, void *data)
+{
+	sym->ptr_die_addr = (uintptr_t)((Dwarf_Die *)data)->addr;
+}
+
+void symbol_set_ptr(struct symbol *sym, Dwarf_Die *ptr)
+{
+	if (for_each(sym->name, set_ptr, ptr) == 0)
 		error("no matching symbols: '%s'", sym->name);
 }