Re: [PATCH] modpost: detect unterminated device id lists

[Andrew Morton <akpm@linux-foundation.org>]

On Tue, 18 Sep 2007 03:15:14 +0530 (IST)
Satyam Sharma <satyam@infradead.org> wrote:

> 
> 
> On Sun, 16 Sep 2007, Andrew Morton wrote:
> 
> > On Mon, 17 Sep 2007 05:54:45 +0530 "Satyam Sharma" <satyam.sharma@gmail.com> wrote:
> > 
> > > On 9/17/07, Andrew Morton <akpm@linux-foundation.org> wrote:
> > > >
> > > > I'm getting this:
> > > >
> > > > rusb2/pvrusb2: struct usb_device_id is 20 bytes.  The last of 3 is:
> > > > 0x03 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00
> > > > 0x00 0x00 0x00 0x00 0x00
> > > > FATAL: drivers/media/video/pvrusb2/pvrusb2: struct usb_device_id is not terminated
> > > > with a NULL entry!
> > > >
> > > > ("rusb2/pvrusb2" ??)
> > > 
> > > Hmm? Are you sure you didn't see any "drivers/media/video/pv" before the
> > > "rusb2/pvrusb2" bit?
> > 
> > Fairly.  I looked twice.
> 
> "drivers/media/video/pvrusb2/pvrusb2" comes out correctly here ...
> 
> 
> > > Looking at Kees' patch (and the existing code), I've no
> > > clue how/why this should happen ... will try to reproduce here ...
> > > 
> > > 
> > > > but:
> > > >
> > > > struct usb_device_id pvr2_device_table[] = {
> > > >         [PVR2_HDW_TYPE_29XXX] = { USB_DEVICE(0x2040, 0x2900) },
> > > >         [PVR2_HDW_TYPE_24XXX] = { USB_DEVICE(0x2040, 0x2400) },
> > > >         { USB_DEVICE(0, 0) },
> > > > };
> > > >
> > > > looks OK?
> > > >
> > > > Using plain old "{ }" shut the warning up.
> > > 
> > > USB_DEVICE(0, 0) is not empty termination, actually, and this looks like
> > > a genuine bug caught by the patch. As that dump shows, USB_DEVICE(0, 0)
> > > assigns "0x03 0x00" (in little endian) to usb_device_id.match_flags. And
> > > I don't think the USB code treats such an entry as an empty entry (?)
> > > 
> > > Interestingly, the "USB_DEVICE(0, 0)" thing is absent from latest -git
> > > tree and also in my copy of 23-rc4-mm1 -- so this looks like something
> > > you must've merged recently.
> > 
> > git-dvb very carefully does
> > 
> > --- a/drivers/media/video/pvrusb2/pvrusb2-hdw.c~git-dvb
> > +++ a/drivers/media/video/pvrusb2/pvrusb2-hdw.c
> > @@ -44,7 +44,7 @@
> >  struct usb_device_id pvr2_device_table[] = {
> >  		[PVR2_HDW_TYPE_29XXX] = { USB_DEVICE(0x2040, 0x2900) },
> >  		[PVR2_HDW_TYPE_24XXX] = { USB_DEVICE(0x2040, 0x2400) },
> > -       { }
> > +       { USB_DEVICE(0, 0) },
> > };
> > 	 
> > MODULE_DEVICE_TABLE(usb, pvr2_device_table);
> 
> Ok, this is a false positive indeed, the core USB code does in fact
> treat such an entry as an empty entry (usb_match_id() tests only the
> .idVendor, .bDeviceClass, .bInterfaceClass and .driver_info members
> for non-zero and not the .match_flags member).
> 
> However, a quick-grep-and-glance tells us that none of the other 2213
> occurrences of USB_DEVICE() in the tree ever do this "(0,0)" thing,
> so it does make sense to change this one to a simple "{ }" as well --
> that's clearer style anyway, and the "standard" way to empty-terminate
> in the rest of the tree, if nothing else.
> 

yeah, I think so.  Mauro, could you please drop that change?