Re: [PATCH v4 03/13] mmc: provide a standard MMC device-tree binding parser centrally

[Simon Horman <horms@verge.net.au>]

On Sat, Feb 16, 2013 at 10:54:36AM +0000, Arnd Bergmann wrote:
> On Saturday 16 February 2013, Simon Horman wrote:
> > > +void mmc_of_parse(struct mmc_host *host)
> > > +{
> > > +     struct device_node *np;
> > > +     u32 bus_width;
> > > +     bool explicit_inv_wp, gpio_inv_wp = false;
> > > +     enum of_gpio_flags flags;
> > > +     int len, ret, gpio;
> > > +
> > > +     if (!host->parent || !host->parent->of_node)
> > > +             return;
> > > +
> > > +     np = host->parent->of_node;
> > > +
> > > +     /* "bus-width" is translated to MMC_CAP_*_BIT_DATA flags */
> > > +     if (of_property_read_u32_array(np, "bus-width", &bus_width, 1) < 0)
> > > +             dev_dbg(host->parent,
> > > +                     "\"bus-width\" property is missing, assuming 1 bit.\n");
> > > +
> > > +     switch (bus_width) {
> > 
> > kbuild tells me that gcc thinks that bus_width is used without being
> > initialised here. Assuming that of_property_read_u32_array always
> > initialises bus_width if it returns zero then perhaps it would be worth
> > considering using uninitialized_var().
> 
> I think this is a false positive that I encountered before and that should be gone
> with gcc-4.7 or higher when using -O2 instead of -Os. I have a patch to disable
> -Wmaybe-uninitialized when builing with -Os. If that gets rid of the warning,
> I'd prefer not annotating it here.
> 
> There was some discussion about removing uninitialized_var() recently after it
> was found to hide some real bugs.

Thanks. I agree that using uninitialized_var() may not be the best idea.