On Mon, 14 Sep 2009, Atsushi Nemoto wrote:
> On Sun, 13 Sep 2009 17:14:06 +0200 (CEST), Julia Lawall <julia@xxxxxxx> wrote:
> > > This patch add some correctness, but obviously incomplete: there are
> > > more error pathes without iounmap/kfree/etc. in this function.
> >
> > The only other error path that I see is:
> >
> > pdev = platform_device_alloc("leds-gpio", basenum);
> > if (!pdev)
> > return;
> >
> > But at that point the call gpiochip_add(&iocled->chip) has already
> > succeeded. From looking at this function, I have the impression that it
> > makes the iocled structure available from a global array, gpio_desc.
> > Since the function containing the above code doesn't return any error
> > code, perhaps the caller will not know whether this platform_device_alloc
> > error occurred or not. There would also be at least the problem of
> > getting the pointer out of the gpio_desc structure. I guess this could be
> > done with gpiochip_remove?
> >
> > I can certainly make a new patch using the goto style, but let me know
> > what to do about the above issue.
>
> Yes, this gpiochip is only used by leds-gpio driver. So
> gpiochip_remove() would be the right thing to do when something
> failed.
>
> Also there is one another error path: platform_device_add() failure at
> the end of this function.
OK, I see. I will submit an improved patch. Thanks for the explanations.
julia
