There are a number of sparse warnings from the latest sparse
snapshot being generated from the drivers/base build. The
main culprits are due to the initialisation functions not
being declared in a header file.
Also, the firmware.c file should include <linux/device.h>
to get the prototype of firmware_register() and
firmware_unregister().
This patch moves the init function declerations from the
init.c file to the base.h, and ensures it is included in
all the relevant c sources. It also adds <linux/device.h>
to the included headers for firmware.c.
The patch does not solve all the sparse errors generated,
but reduces the count significantly.
drivers/base/core.c:161:1: warning: symbol 'devices_subsys' was not declared. Should it be static?
drivers/base/core.c:417:12: warning: symbol 'devices_init' was not declared. Should it be static?
drivers/base/sys.c:253:6: warning: symbol 'sysdev_shutdown' was not declared. Should it be static?
drivers/base/sys.c:326:5: warning: symbol 'sysdev_suspend' was not declared. Should it be static?
drivers/base/sys.c:428:5: warning: symbol 'sysdev_resume' was not declared. Should it be static?
drivers/base/sys.c:450:12: warning: symbol 'system_bus_init' was not declared. Should it be static?
drivers/base/bus.c:133:1: warning: symbol 'bus_subsys' was not declared. Should it be static?
drivers/base/bus.c:667:12: warning: symbol 'buses_init' was not declared. Should it be static?
drivers/base/class.c:759:12: warning: symbol 'classes_init' was not declared. Should it be static?
drivers/base/platform.c:313:12: warning: symbol 'platform_bus_init' was not declared. Should it be static?
drivers/base/cpu.c:110:12: warning: symbol 'cpu_dev_init' was not declared. Should it be static?
drivers/base/firmware.c:17:5: warning: symbol 'firmware_register' was not declared. Should it be static?
drivers/base/firmware.c:23:6: warning: symbol 'firmware_unregister' was not declared. Should it be static?
drivers/base/firmware.c:28:12: warning: symbol 'firmware_init' was not declared. Should it be static?
drivers/base/init.c:28:13: warning: symbol 'driver_init' was not declared. Should it be static?
drivers/base/dmapool.c:174:10: warning: implicit cast from nocast type
drivers/base/attribute_container.c:439:1: warning: symbol 'attribute_container_init' was not declared. Should it be static?
drivers/base/power/runtime.c:76:6: warning: symbol 'dpm_set_power_state' was not declared. Should it be static?
Signed-off-by: Ben Dooks <[email protected]>
On Thu, 13 Oct 2005, Ben Dooks wrote:
>
> The patch does not solve all the sparse errors generated,
> but reduces the count significantly.
Well, you should also then remove the _bad_ declarations.
For example, attribute_container_init() right now is defined in
attribute_container.c, but then it's _declared_ (with no checking) where
it's used in init.c.
The sparse warnign is appropriate: it was not declared where that
declaration is actually visible to the definition, so the code basically
isn't type-safe at all (since there's nothing that enforces the
declaration actually matching the definition).
You made the declaration properly visible, but you should also remove the
bogus declaration. A declaration that isn't visible to the definition is
always bad - since in the absense of a compiler with global visibility it
may or may not actually match what it supposedly declares.
I wonder if I should make sparse warn about multiple declarations..
These days, sparse actually has some limited support for checking _global_
visibility, and we could do cross-checking across thousands of files.
However, the build environment isn't really very amenable to that, so
doing a global sparse check is pretty hard in practice.
We could possibly do a per-directory global check, which might be better
than nothing (ie if you were to have incorrect declaration in a C file
that is in the same directory as another C file, then we could
cross-check).
But the kernel kbuild environment is pretty hairy, and I wouldn't even
know where to begin trying to do that. It's also fundamentally hard to do
if there are per-file pre-defines (since to do a cross-check, sparse wants
to see all C files together on the command line).
Linus
On Thu, Oct 13, 2005 at 11:10:15AM -0700, Linus Torvalds wrote:
> On Thu, 13 Oct 2005, Ben Dooks wrote:
> >
> > The patch does not solve all the sparse errors generated,
> > but reduces the count significantly.
>
> Well, you should also then remove the _bad_ declarations.
>
> ...
>
> You made the declaration properly visible, but you should also remove the
> bogus declaration. A declaration that isn't visible to the definition is
> always bad - since in the absense of a compiler with global visibility it
> may or may not actually match what it supposedly declares.
Erm, lets take your example - attribute_container_init(). It's defined
in attribute_container.c, where the base.h include was added:
diff -urpN -X ../dontdiff linux-2.6.14-rc4-bjd1/drivers/base/attribute_container.c linux-2.6.14-rc4-bjd2/drivers/base/attribute_container.c
--- linux-2.6.14-rc4-bjd1/drivers/base/attribute_container.c 2005-10-11 10:56:31.000000000 +0100
+++ linux-2.6.14-rc4-bjd2/drivers/base/attribute_container.c 2005-10-13 15:30:01.000000000 +0100
@@ -19,6 +19,8 @@
#include <linux/list.h>
#include <linux/module.h>
+#include "base.h"
+
/* This is a private structure used to tie the classdev and the
* container .. it should never be visible outside this file */
struct internal_container {
The base.h include contains the definition:
diff -urpN -X ../dontdiff linux-2.6.14-rc4-bjd1/drivers/base/base.h linux-2.6.14-rc4-bjd2/drivers/base/base.h
--- linux-2.6.14-rc4-bjd1/drivers/base/base.h 2005-09-01 21:02:36.000000000 +0100
+++ linux-2.6.14-rc4-bjd2/drivers/base/base.h 2005-10-13 15:26:56.000000000 +0100
@@ -1,3 +1,15 @@
+
+/* initialisation functions */
+
+extern int devices_init(void);
+extern int buses_init(void);
+extern int classes_init(void);
+extern int firmware_init(void);
+extern int platform_bus_init(void);
+extern int system_bus_init(void);
+extern int cpu_dev_init(void);
+extern int attribute_container_init(void);
+
extern int bus_add_device(struct device * dev);
extern void bus_remove_device(struct device * dev);
And base.h was included in init.c and the bogus declaration removed:
diff -urpN -X ../dontdiff linux-2.6.14-rc4-bjd1/drivers/base/init.c linux-2.6.14-rc4-bjd2/drivers/base/init.c
--- linux-2.6.14-rc4-bjd1/drivers/base/init.c 2005-06-17 20:48:29.000000000 +0100
+++ linux-2.6.14-rc4-bjd2/drivers/base/init.c 2005-10-13 15:27:05.000000000 +0100
@@ -10,14 +10,8 @@
#include <linux/device.h>
#include <linux/init.h>
-extern int devices_init(void);
-extern int buses_init(void);
-extern int classes_init(void);
-extern int firmware_init(void);
-extern int platform_bus_init(void);
-extern int system_bus_init(void);
-extern int cpu_dev_init(void);
-extern int attribute_container_init(void);
+#include "base.h"
+
/**
* driver_init - initialize driver model.
*
I can't see anything that was missed.
--
Russell King
Linux kernel 2.6 ARM Linux - http://www.arm.linux.org.uk/
maintainer of: 2.6 Serial core
On Thu, Oct 13, 2005 at 11:10:15AM -0700, Linus Torvalds wrote:
>
>
> On Thu, 13 Oct 2005, Ben Dooks wrote:
> >
> > The patch does not solve all the sparse errors generated,
> > but reduces the count significantly.
>
> Well, you should also then remove the _bad_ declarations.
Sorry, I do not follow you, can you clarify this for me.
My patch did not generate any more errors, just removed the ones
that where easy to see a solution too.
> For example, attribute_container_init() right now is defined in
> attribute_container.c, but then it's _declared_ (with no checking) where
> it's used in init.c.
>
> The sparse warnign is appropriate: it was not declared where that
> declaration is actually visible to the definition, so the code basically
> isn't type-safe at all (since there's nothing that enforces the
> declaration actually matching the definition).
>
> You made the declaration properly visible, but you should also remove the
> bogus declaration. A declaration that isn't visible to the definition is
> always bad - since in the absense of a compiler with global visibility it
> may or may not actually match what it supposedly declares.
>
> I wonder if I should make sparse warn about multiple declarations..
I pulled the old declerations out of drivers/base/init.c ? I'm sure
the patch shows that?
--
Ben ([email protected], http://www.fluff.org/)
'a smiley only costs 4 bytes'
On Thu, 13 Oct 2005, Russell King wrote:
>
> Erm, lets take your example - attribute_container_init(). It's defined
> in attribute_container.c, where the base.h include was added:
>
> I can't see anything that was missed.
Duh. My mistake, I looked at the patch twice, but I still missed the place
where it removed the declaration. Twice.
Gaah. Where are my brain-pills again?
Linus