Bug 14411

Summary:	[FR] automatic kernel module check for symbol-related problems
Product:	Sisyphus	Reporter:	Ivan Zakharyaschev <imz>
Component:	rpm-build-kernel	Assignee:	Sergey Vlasov <vsu>
Status:	CLOSED FIXED	QA Contact:	qa-sisyphus
Severity:	enhancement
Priority:	P2	CC:	boyarsh, glebfm, iv, kernelbot, ldv, mike, rider, sbolshakov, shrek, sin, vitty, vsu, vt, zerg
Version:	unstable
Hardware:	all
OS:	Linux
Bug Depends on:
Bug Blocks:	14408

Description

Ivan Zakharyaschev 2008-02-13 11:18:36 MSK

rpm-build-kernel-0.11-alt1

After modules have been compiled, it would be nice to check if they can be
loaded into the corresponding kernel, whether they have no unknown symbols.
Before it happens in a working system.

Perhaps, module-init-tools are able to do such a check (not for the running
kernel), or it could be done in a hardware emulator.

This feature request was inspired by
https://bugzilla.altlinux.org/show_bug.cgi?id=14408 ,
https://bugzilla.altlinux.org/show_bug.cgi?id=14409 .

Comment 1

Sergey Vlasov 2008-02-13 13:30:51 MSK

The kernel build system already does some checks for undefined symbols;
unfortunately, even in 2.6.24 failures of these checks when building external
modules are just warnings.

However, the real problem with drm is not undefined symbols, but multiple
definitions of symbols (the drm module cannot be loaded due to the duplicate
idr_remove_all definition, and then all modules which depend on it fail with
unresolved symbols), and the uhci_hcd problem is due to races between concurrent
modprobe invocations.

Comment 2

Ivan Zakharyaschev 2008-02-13 13:57:23 MSK

(In reply to comment #1)

> However, the real problem with drm is not undefined symbols, but multiple
> definitions of symbols (the drm module cannot be loaded due to the duplicate
> idr_remove_all definition,

I see. Added "multiple definitions of symbols" to the summary.

Comment 3

Michael Shigorin 2009-09-13 17:08:19 MSD

Guess VM support is out of rpm-build-kernel scope.

Comment 4

Anton Farygin 2026-09-10 08:06:28 MSK

Проверка, о которой просили, теперь делается самой сборочной системой ядра, и не предупреждением, а ошибкой. Собрал в контейнере Sisyphus внешний модуль с заведомо неразрешённым символом против kernel-headers-modules-6.12:

	ERROR: modpost: "this_symbol_does_not_exist" [/tmp/m/hello.ko] undefined!
	make[2]: *** [scripts/Makefile.modpost:145: /tmp/m/Module.symvers] Error 1
	make rc=2

Файл .ko не создаётся, сборка падает — модуль с неразрешёнными символами в репозиторий не попадёт. Дублирующиеся экспорты modpost тоже ловит («exports duplicate symbol»). Проверка загрузки собранных модулей в виртуальной машине, как и говорил mike@ в комментарии 3, выходит за рамки rpm-build-kernel (сейчас это бинарный пакет из kernel-build-tools 0.132-alt1). FIXED.