Thread (35 messages) 35 messages, 3 authors, 14d ago

Re: [PATCH v2 19/24] dyndbg,module: make proper substructs in _ddebug_info

From: sashiko-bot@kernel.org
Date: 2026-05-23 07:45:09

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:
- [High] Module loader searches for incorrectly named dynamic debug sections, completely disabling dynamic debug for all loadable modules.
- [High] Built-in modules without class maps erroneously inherit all global system-wide class maps due to missing state reset.
- [Low] Renaming `struct ddebug_class_param` breaks compilation of external users like the DRM subsystem.
--

commit 647ef6a50774bbbe7f638336c42ae9d912b4ca12
Author: Jim Cromie [off-list ref]

dyndbg,module: make proper substructs in _ddebug_info

This commit recomposes struct _ddebug_info by moving the descriptors and class
maps fields into new sub-structures struct _ddebug_descs and struct
_ddebug_class_maps. It also normalizes all related struct names to use a
_ddebug_ prefix.
quoted hunk ↗ jump to hunk
diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h
index a84dcc127e5a7..001bc84eadcab 100644
--- a/include/linux/dynamic_debug.h
+++ b/include/linux/dynamic_debug.h
[ ... ]
quoted hunk ↗ jump to hunk
@@ -81,21 +81,33 @@ struct ddebug_class_map {
 	enum ddebug_class_map_type map_type;
 };
 
-/* encapsulate linker provided built-in (or module) dyndbg data */
+/*
+ * @_ddebug_info: gathers module/builtin dyndbg_* __sections together.
+ * For builtins, it is used as a cursor, with the inner structs
+ * marking sub-vectors of the builtin __sections in DATA.
+ */
+struct _ddebug_descs {
+	struct _ddebug *start;
+	unsigned int len;
+};
+
+struct _ddebug_class_maps {
+	struct _ddebug_class_map *start;
+	unsigned int len;
+};
+
 struct _ddebug_info {
-	struct _ddebug *descs;
-	struct ddebug_class_map *classes;
-	unsigned int num_descs;
-	unsigned int num_classes;
+	struct _ddebug_descs descs;
+	struct _ddebug_class_maps maps;
 };
 
-struct ddebug_class_param {
+struct _ddebug_class_param {
[Severity: Low]
Does renaming this structure break compilation for external users?

For example, the DRM subsystem directly references the old structure name in
drivers/gpu/drm/drm_print.c:
"static struct ddebug_class_param drm_debug_bitmap"

Failing to update these external users will result in a build failure for
configurations enabling DRM.
 	union {
 		u64 *bits;
 		u64 *lvl;
 	};
 	char flags[8];
-	const struct ddebug_class_map *map;
+	const struct _ddebug_class_map *map;
 };
 
 /*
[ ... ]
quoted hunk ↗ jump to hunk
diff --git a/kernel/module/main.c b/kernel/module/main.c
index 46dd8d25a6058..c2b6e70f2e773 100644
--- a/kernel/module/main.c
+++ b/kernel/module/main.c
@@ -2774,12 +2774,12 @@ static int find_module_sections(struct module *mod, struct load_info *info)
 		pr_warn("%s: Ignoring obsolete parameters\n", mod->name);
 
 #ifdef CONFIG_DYNAMIC_DEBUG_CORE
-	mod->dyndbg_info.descs = section_objs(info, "__dyndbg",
-					      sizeof(*mod->dyndbg_info.descs),
-					      &mod->dyndbg_info.num_descs);
-	mod->dyndbg_info.classes = section_objs(info, "__dyndbg_classes",
-						sizeof(*mod->dyndbg_info.classes),
-						&mod->dyndbg_info.num_classes);
+	mod->dyndbg_info.descs.start = section_objs(info, "__dyndbg_descriptors",
+						    sizeof(*mod->dyndbg_info.descs.start),
+						    &mod->dyndbg_info.descs.len);
+	mod->dyndbg_info.maps.start = section_objs(info, "__dyndbg_class_maps",
+						   sizeof(*mod->dyndbg_info.maps.start),
+						   &mod->dyndbg_info.maps.len);
[Severity: High]
Does this correctly match the section names emitted by the linker?

It looks like the section names here were updated to "__dyndbg_descriptors"
and "__dyndbg_class_maps", but the macros like DECLARE_DYNDBG_CLASSMAP and
DEFINE_DYNAMIC_DEBUG_METADATA still emit to "__dyndbg" and "__dyndbg_classes".

Because the strings do not match, section_objs() will fail to find the
sections, setting their lengths to 0 and effectively discarding all
dynamic debug prints for loadable modules.

I see this regression is fixed later in the series by commit f9409a020a4bc
("dyndbg: improve section names").
 #endif
 
 	return 0;
[ ... ]
quoted hunk ↗ jump to hunk
diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index 580aa258f902f..169b94ff6f8ed 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
[ ... ]
quoted hunk ↗ jump to hunk
@@ -1202,10 +1203,10 @@ static int ddebug_add_module(struct _ddebug_info *di, const char *modname)
 {
 	struct ddebug_table *dt;
 
-	if (!di->num_descs)
+	if (!di->descs.len)
 		return 0;
 
-	v3pr_info("add-module: %s %d sites\n", modname, di->num_descs);
+	v3pr_info("add-module: %s %d sites\n", modname, di->descs.len);
 
 	dt = kzalloc_obj(*dt);
 	if (dt == NULL) {
@@ -1219,19 +1220,18 @@ static int ddebug_add_module(struct _ddebug_info *di, const char *modname)
 	 * this struct ddebug_table.
 	 */
 	dt->mod_name = modname;
-	dt->ddebugs = di->descs;
-	dt->num_ddebugs = di->num_descs;
+	dt->info = *di;
[Severity: High]
Does this code leave dt->info.maps initialized to the entire system-wide
class maps array if the built-in module has no class maps?

In dynamic_debug_init(), the global _ddebug_info di is initialized with
.maps spanning the entire __dyndbg_classes section.

When this is copied via dt->info = *di here, dt->info.maps also spans
the entire section.

If the module has no class maps, ddebug_attach_module_classes() will run, but
the local variable nc remains 0. The block if (nc) at the end is skipped,
and dt->info.maps is never updated to reflect 0 length.

As a result, built-in modules without class maps falsely match unrelated
class maps from other built-in modules.

I see this regression is fixed later in the series by commit 6fc0ea29f4f01
("dyndbg: hoist classmap-filter-by-modname up to ddebug_add_module").
 
 	INIT_LIST_HEAD(&dt->link);
 
-	if (di->classes && di->num_classes)
+	if (di->maps.len)
 		ddebug_attach_module_classes(dt, di);
 
 	mutex_lock(&ddebug_lock);
-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260523-dd-maint-2-v2-0-b937312aa083@gmail.com?part=19
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help