[PATCH] fail kernel compilation in case of unresolved symbols (v2)

[PATCH] fail kernel compilation in case of unresolved symbols (v2)

[Kirill Korotaev]

At stage 2 modpost utility is used to check modules.
In case of unresolved symbols modpost only prints warning.

IMHO it is a good idea to fail compilation process in case of
unresolved symbols (at least in modules coming with kernel),
since usually such errors are left unnoticed, but kernel
modules are broken.

Changes from v1:
- new option '-w' is added to modpost:
  if option is specified, modpost only warns about unresolved symbols
- modpost is called with '-w' for external modules in Makefile.modpost

Signed-Off-By: Andrey Mirkin <amirkin@sw.ru>
Signed-Off-By: Kirill Korotaev <dev@openvz.org>

---

diff --git a/scripts/Makefile.modpost b/scripts/Makefile.modpost
index 0a64688..9c01886 100644
--- a/scripts/Makefile.modpost
+++ b/scripts/Makefile.modpost
@@ -58,6 +58,7 @@ quiet_cmd_modpost = MODPOST
 	$(if $(KBUILD_EXTMOD),-i,-o) $(kernelsymfile) \
 	$(if $(KBUILD_EXTMOD),-I $(modulesymfile)) \
 	$(if $(KBUILD_EXTMOD),-o $(modulesymfile)) \
+	$(if $(KBUILD_EXTMOD),-w) \
 	$(filter-out FORCE,$^)
 
 PHONY += __modpost
diff --git a/scripts/mod/modpost.c b/scripts/mod/modpost.c
index dfde0e8..083a75e 100644
--- a/scripts/mod/modpost.c
+++ b/scripts/mod/modpost.c
@@ -23,6 +23,8 @@ int have_vmlinux = 0;
 static int all_versions = 0;
 /* If we are modposting external module set to 1 */
 static int external_module = 0;
+/* Only warn about unresolved symbols */
+static int warn_unresolved = 0;
 /* How a symbol is exported */
 enum export {
 	export_plain,      export_unused,     export_gpl,
@@ -1187,16 +1189,19 @@ static void add_header(struct buffer *b,
 /**
  * Record CRCs for unresolved symbols
  **/
-static void add_versions(struct buffer *b, struct module *mod)
+static int add_versions(struct buffer *b, struct module *mod)
 {
 	struct symbol *s, *exp;
+	int err = 0;
 
 	for (s = mod->unres; s; s = s->next) {
 		exp = find_symbol(s->name);
 		if (!exp || exp->module == mod) {
-			if (have_vmlinux && !s->weak)
+			if (have_vmlinux && !s->weak) {
 				warn("\"%s\" [%s.ko] undefined!\n",
 				     s->name, mod->name);
+				err = warn_unresolved ? 0 : 1;
+			}
 			continue;
 		}
 		s->module = exp->module;
@@ -1205,7 +1210,7 @@ static void add_versions(struct buffer *
 	}
 
 	if (!modversions)
-		return;
+		return err;
 
 	buf_printf(b, "\n");
 	buf_printf(b, "static const struct modversion_info ____versions[]\n");
@@ -1225,6 +1230,8 @@ static void add_versions(struct buffer *
 	}
 
 	buf_printf(b, "};\n");
+
+	return err;
 }
 
 static void add_depends(struct buffer *b, struct module *mod,
@@ -1402,8 +1409,9 @@ int main(int argc, char **argv)
 	char *kernel_read = NULL, *module_read = NULL;
 	char *dump_write = NULL;
 	int opt;
+	int err;
 
-	while ((opt = getopt(argc, argv, "i:I:mo:a")) != -1) {
+	while ((opt = getopt(argc, argv, "i:I:mo:aw")) != -1) {
 		switch(opt) {
 			case 'i':
 				kernel_read = optarg;
@@ -1421,6 +1429,9 @@ int main(int argc, char **argv)
 			case 'a':
 				all_versions = 1;
 				break;
+			case 'w':
+				warn_unresolved = 1;
+				break;
 			default:
 				exit(1);
 		}
@@ -1441,6 +1452,8 @@ int main(int argc, char **argv)
 		check_exports(mod);
 	}
 
+	err = 0;
+
 	for (mod = modules; mod; mod = mod->next) {
 		if (mod->skip)
 			continue;
@@ -1448,7 +1461,7 @@ int main(int argc, char **argv)
 		buf.pos = 0;
 
 		add_header(&buf, mod);
-		add_versions(&buf, mod);
+		err |= add_versions(&buf, mod);
 		add_depends(&buf, mod, modules);
 		add_moddevtable(&buf, mod);
 		add_srcversion(&buf, mod);
@@ -1460,5 +1473,5 @@ int main(int argc, char **argv)
 	if (dump_write)
 		write_dump(dump_write);
 
-	return 0;
+	return err;
 }

Re: [PATCH] fail kernel compilation in case of unresolved symbols (v2)

[Olaf Hering]

On Thu, Sep 07, Kirill Korotaev wrote:

> At stage 2 modpost utility is used to check modules.
> In case of unresolved symbols modpost only prints warning.
> 
> IMHO it is a good idea to fail compilation process in case of
> unresolved symbols (at least in modules coming with kernel),
> since usually such errors are left unnoticed, but kernel
> modules are broken.

It clearly depends on the context. An unimportant dvb module may have
unresolved symbols, but the drivers for your root filesystem should
rather not have unresolved symbols.

Better leave the current default, your patch seems to turn an unresolved
symbol with "unknown importance" into a hard error.

Re: [PATCH] fail kernel compilation in case of unresolved symbols (v2)

[Adrian Bunk]

On Thu, Sep 07, 2006 at 01:05:13PM +0200, Olaf Hering wrote:
> On Thu, Sep 07, Kirill Korotaev wrote:
> 
> > At stage 2 modpost utility is used to check modules.
> > In case of unresolved symbols modpost only prints warning.
> > 
> > IMHO it is a good idea to fail compilation process in case of
> > unresolved symbols (at least in modules coming with kernel),
> > since usually such errors are left unnoticed, but kernel
> > modules are broken.
> 
> It clearly depends on the context. An unimportant dvb module may have
> unresolved symbols, but the drivers for your root filesystem should
> rather not have unresolved symbols.
> 
> Better leave the current default, your patch seems to turn an unresolved
> symbol with "unknown importance" into a hard error.

If any module shipped with the kernel has in any configuration 
unresolved symbols that's a bug that should be reported, not ignored.

And changing runtime errors to build errors ensures that such errors 
never reach users (if the module is really unimportant disabling it 
is easy).

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed

Re: [PATCH] fail kernel compilation in case of unresolved symbols (v2)

[Olaf Hering]

On Thu, Sep 07, Adrian Bunk wrote:

> If any module shipped with the kernel has in any configuration 
> unresolved symbols that's a bug that should be reported, not ignored.

Yes, but on request when building the package. Not per default.
I probably missed the reason why this is now suddenly a problem.

Re: [PATCH] fail kernel compilation in case of unresolved symbols (v2)

[Kirill Korotaev]

Olaf Hering wrote:
> On Thu, Sep 07, Adrian Bunk wrote:
> 
> 
>>If any module shipped with the kernel has in any configuration 
>>unresolved symbols that's a bug that should be reported, not ignored.
> 
> 
> Yes, but on request when building the package. Not per default.
> I probably missed the reason why this is now suddenly a problem.
It is not that sudden at all. I experienced this problem many times so far
and working with a build system came to the idea of failing
builds when there are unresolved symbols.

I'm pretty sure that having this patch in mainstream
will make unresolved symbols a rare problem as many of them will be fixed soon.
So I'm pretty agree with Adrian that modules with unresolved symbols is a bug
and it MUST be fixed.
I would be very much interested to hear Andrew opinion on this as
he probably makes kernels even more often than any of us :)

Thanks,
Kirill

Re: [PATCH] fail kernel compilation in case of unresolved symbols (v2)

[Andrew Morton]

On Thu, 07 Sep 2006 19:23:49 +0400
Kirill Korotaev <dev@sw.ru> wrote:

> Olaf Hering wrote:
> > On Thu, Sep 07, Adrian Bunk wrote:
> > 
> > 
> >>If any module shipped with the kernel has in any configuration 
> >>unresolved symbols that's a bug that should be reported, not ignored.
> > 
> > 
> > Yes, but on request when building the package. Not per default.
> > I probably missed the reason why this is now suddenly a problem.
> It is not that sudden at all. I experienced this problem many times so far
> and working with a build system came to the idea of failing
> builds when there are unresolved symbols.
> 
> I'm pretty sure that having this patch in mainstream
> will make unresolved symbols a rare problem as many of them will be fixed soon.
> So I'm pretty agree with Adrian that modules with unresolved symbols is a bug
> and it MUST be fixed.
> I would be very much interested to hear Andrew opinion on this as
> he probably makes kernels even more often than any of us :)
> 

Am sympathetic to the idea.

Some architectures (eg sparc64) generate large numbers of unresolved module
symbol warnings during an allmodconfig build.  It's not a big problem in
practice - these are subsystems which everyone statically links anyway.

But it does mean that additional work needs to be done if we want to
prevent these allmodconfig builds from erroring out.

Re: [PATCH] fail kernel compilation in case of unresolved symbols (v2)

[Chris Wedgwood]

On Thu, Sep 07, 2006 at 02:03:09PM +0400, Kirill Korotaev wrote:

> At stage 2 modpost utility is used to check modules.  In case of
> unresolved symbols modpost only prints warning.

please don't, i get bogus warnings for some drivers when
cross-compiling, this would create problems for little or no gain

Re: [PATCH] fail kernel compilation in case of unresolved symbols (v2)

[Adrian Bunk]

On Thu, Sep 07, 2006 at 05:13:25PM -0700, Chris Wedgwood wrote:
> On Thu, Sep 07, 2006 at 02:03:09PM +0400, Kirill Korotaev wrote:
> 
> > At stage 2 modpost utility is used to check modules.  In case of
> > unresolved symbols modpost only prints warning.
> 
> please don't, i get bogus warnings for some drivers when
> cross-compiling, this would create problems for little or no gain

Can you give an example of such a bogus warning?

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed