Re: [Xen-devel] [PATCH 1/2] build/clang: remove the address-of-packed-member warning

[Tim Deegan <tim@xxxxxxx>]

At 14:36 +0000 on 06 Mar (1488811016), George Dunlap wrote:
> On 06/03/17 13:58, Jan Beulich wrote:
> >>>> On 06.03.17 at 13:31, <roger.pau@xxxxxxxxxx> wrote:
> >> --- a/Config.mk
> >> +++ b/Config.mk
> >> @@ -216,6 +216,7 @@ $(call 
> >> cc-option-add,HOSTCFLAGS,HOSTCC,-Wdeclaration-after-statement)
> >>  $(call cc-option-add,CFLAGS,CC,-Wdeclaration-after-statement)
> >>  $(call cc-option-add,CFLAGS,CC,-Wno-unused-but-set-variable)
> >>  $(call cc-option-add,CFLAGS,CC,-Wno-unused-local-typedefs)
> >> +$(call cc-option-add,CFLAGS,CC,-Wno-address-of-packed-member)
> > 
> > Actually, having thought some more about this, the warning
> > should be suppressed only for x86 imo. ARM wants aligned
> > accesses after all.
> 
> Looking at Roger's complaint, it appears that the warning is issued even
> if the member actually is aligned, if *on some unknown system*, it might
> someday be un-aligned.

AIUI the complaint is (based on the simplified example from the ticket):

    struct __attribute__((__packed__)) bar {
        uint16_t x1;
        uint16_t x2;
    } b;
    
    &b.x2;

Because the struct is packed, it has alignment 1, and so do its
fields.   &b.x2 is a pointer to a uint16_t, but it _isn't_ 16-bit
aligned (because the whole struct is only byte-aligned).

So I guess that one fix would be to declare that the struct has
appropriate alignment?  I don't know whether that would suppress the
warning, but the clang devs might be more receptive to seeing it as
a false positive.

Tim.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel