From: Alan Stern on
Fake "address-of" expressions that evaluate to NULL generally confuse
readers and can provoke compiler warnings. This patch (as1411)
removes one such fake expression, using an "#ifdef" in its place.

Signed-off-by: Alan Stern <stern(a)>


On Wed, 4 Aug 2010, Linus Torvalds wrote:

> On Tue, Aug 3, 2010 at 9:13 PM, James Bottomley <James.Bottomley(a)> wrote:
> >
> > Alan Stern (3):
> > � � �sd: add support for runtime PM
> > � � �implement runtime Power Management
> > � � �convert to the new PM framework
> Guys, these kind of crazy games really aren't appropriate:
> +/* scsi_pm.c */
> +#ifdef CONFIG_PM_OPS
> +extern const struct dev_pm_ops scsi_bus_pm_ops;
> +#else
> +#define scsi_bus_pm_ops (*NULL)
> +#endif
> that's just crazy. Yes, I see how it's then used (address-of operator
> turns it back into NULL), but the compiler warns about it
> ("drivers/scsi/scsi_sysfs.c:384: warning: dereferencing �void *�
> pointer") and I think the compiler is 100% correct about warning about
> it.
> It's not just the (*NULL) games, btw. The above can cause confusion.
> It's ugly not just because it causes the compiler to warn, but because
> you use a very subtle and non-standard way of using #define's, so that
> when you look at the source code where this is used, it's not at all
> obvious what is going on. The code looks like
> .pm = &scsi_bus_pm_ops,
> and dammit, it would be rather understandable if some _human_ that
> reads that is also confused and thinks that the above means that the
> .pm pointer can never be NULL. The address-of would certainly throw
> me, and not necessarily at all make me grep for "could that possibly
> be some crazy way to say NULL".
> And there is absolutely no reason to play games like that. It would
> have been entirely understandable if you just put the #ifdef in the C
> code itself, or if you used a #define that just said
> #ifdef CONFIG_PM_OPS
> #define SCSI_BUS_PM_OPS &scsi_bus_pm_ops
> #else
> #endif
> and I think it would be less confusing, and it wouldn't upset the compiler.
> Yes, yes, I realize that we do these kinds of things for function
> pointers all the time, so I do understand where the pattern comes
> from. At the same time, I rather think that function pointers are a
> bit different, and they don't have the whole address-of problem.
> I guess I should be happy that you didn't use some linker tricks to
> make "&scsi_bus_pm_ops" turn into NULL at link time. That could be
> done too, and would have been even more subtly confusing.

In my opinion this is the simplest and easiest-to-understand approach.
The two of you can fight over who merges it first. :-)

Alan Stern

Index: usb-2.6/drivers/scsi/scsi_priv.h
--- usb-2.6.orig/drivers/scsi/scsi_priv.h
+++ usb-2.6/drivers/scsi/scsi_priv.h
@@ -148,8 +148,6 @@ static inline void scsi_netlink_exit(voi
/* scsi_pm.c */
extern const struct dev_pm_ops scsi_bus_pm_ops;
-#else /* CONFIG_PM_OPS */
-#define scsi_bus_pm_ops (*NULL)
extern void scsi_autopm_get_target(struct scsi_target *);
Index: usb-2.6/drivers/scsi/scsi_sysfs.c
--- usb-2.6.orig/drivers/scsi/scsi_sysfs.c
+++ usb-2.6/drivers/scsi/scsi_sysfs.c
@@ -382,7 +382,9 @@ struct bus_type scsi_bus_type = {
.name = "scsi",
.match = scsi_bus_match,
.uevent = scsi_bus_uevent,
.pm = &scsi_bus_pm_ops,

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)
More majordomo info at
Please read the FAQ at