On Fri, Nov 15, 2013 at 03:56:31PM -0000, David Laight wrote: > > From: Hans de Goede [mailto:hdegoede@xxxxxxxxxx] > > On 11/15/2013 04:29 PM, David Laight wrote: > > >> From: Of Hans de Goede > > >> > > >> The iu struct definitions are usb packet definitions, so no alignment should > > >> happen. Notice that assuming 32 bit alignment this does not make any > > >> difference at all. > > >> > > >> Signed-off-by: Hans de Goede <hdegoede@xxxxxxxxxx> > > >> --- > > >> include/linux/usb/uas.h | 10 +++++----- > > >> 1 file changed, 5 insertions(+), 5 deletions(-) > > >> > > >> diff --git a/include/linux/usb/uas.h b/include/linux/usb/uas.h > > >> index 772b66b..3fc8e8b 100644 > > >> --- a/include/linux/usb/uas.h > > >> +++ b/include/linux/usb/uas.h > > >> @@ -9,7 +9,7 @@ struct iu { > > >> __u8 iu_id; > > >> __u8 rsvd1; > > >> __be16 tag; > > >> -}; > > >> +} __attribute__((__packed__)); > > > > > > Won't this cause the compiler to generate multiple byte accesses > > > for the 'tag' field in systems that don't support misaligned > > > accesses. > > > > The tag field is aligned to a multiple of its size, so that should not > > happen, unless some system wants 32 bits alignment for 16 bit variables. > > That isn't going to be true. It might even be against the C standard. > In any case such a system wouldn't be able to define __be16 > (unless the compiler implemented it using byte accesses). > > ... > > Without packed struct members will never be mis-aligned in memory as the > > compiler will simply add padding. packed is used when the struct represent > > data coming from an outside source, ie a file or a socket, or in this case > > an usb data packet. > > What would be useful is an attribute that errorred if any implicit padding > were added. > > If you are worried about padding add a compile-time test on the size of the > structure. No, Linux relies on the compiler to pack things like this together and not put padding in, with the __packed__ attribute (don't we have a __packed macro instead of the __attribute__() thing?) So this is correct, if it changes, then huge swatches of Linux kernel code will break. Oh, and gcc, llvm, and icc all do things this way, and any new c compiler will do it as well if they want to build the kernel. greg k-h -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html