On Mon, Apr 20, 2020 at 7:28 AM Masahiro Yamada <masahiroy@xxxxxxxxxx> wrote: > > As the bug report [1] pointed out, <linux/vermagic.h> must be included > after <linux/module.h>. > > I believe we should be able to include headers in any order. We often > sort include directives alphabetically, but it is just coding style > convention. Technically, we can get rid of the include order restriction > by making every header self-contained. > > Currently, arch-specific MODULE_ARCH_VERMAGIC is defined in > <asm/module.h>, but it is not included from <linux/vermagic.h>. > > Hence, the straight-forward fix-up would be as follows: > > |--- a/include/linux/vermagic.h > |+++ b/include/linux/vermagic.h > |@@ -1,5 +1,6 @@ > | /* SPDX-License-Identifier: GPL-2.0 */ > | #include <generated/utsrelease.h> > |+#include <linux/module.h> > | > | /* Simply sanity version stamp for modules. */ > | #ifdef CONFIG_SMP > > This works enough, but for further cleanups, I split MODULE_ARCH_VERMAGIC > definitions into <asm/vermagic.h>. > > With this, <linux/module.h> and <linux/vermagic.h> will be orthogonal, > and the location of MODULE_ARCH_VERMAGIC definitions will be consistent. > > For arc and ia64, MODULE_PROC_FAMILY is only used for defining > MODULE_ARCH_VERMAGIC. I squashed it. > > FOR hexagon, nds32, and xtensa, I removed <asm/modules.h> entirely > because they contained nothing but MODULE_ARCH_VERMAGIC definition. > Kbuild will automatically generate <asm/modules.h> at build-time, > wrapping <asm-generic/module.h>. > > [1] https://lore.kernel.org/lkml/20200411155623.GA22175@xxxxxxx > I forgot to add this: Reported-by: Borislav Petkov <bp@xxxxxxx> > Signed-off-by: Masahiro Yamada <masahiroy@xxxxxxxxxx> > --- > > I do not mean to replace the work by Leon Romanovsky: > https://lkml.org/lkml/2020/4/19/201 > > His work intends to hide <linux/vermagic.h> from driver writers. > It is solving a different problem. > It does not solve the include order restriction reported by [1]. > It still relies on kernel/module.c and *.mod.c > include <linux/vermagic.h> after <linux/module.h>. > > I believe we should not impose any restriction about the include order. > So, this patch is the direct answer to the bug report [1]. > > BTW, I think commit f58dd03b1157bdf3b64c36e9525f8d7f69c25df2 > was a bad way to suppress the problem, but that is another story. > -- Best Regards Masahiro Yamada
