On Fri, Jul 12, 2019 at 3:54 PM Roman Kagan <rkagan@xxxxxxxxxxxxx> wrote:
>
> On Fri, Jul 12, 2019 at 03:32:43PM +0200, Arnd Bergmann wrote:
> > clang points out that running a 64-bit guest on a 32-bit host
> > would lead to uninitialized variables:
> >
> > arch/x86/kvm/hyperv.c:1610:6: error: variable 'ingpa' is used uninitialized whenever 'if' condition is false [-Werror,-Wsometimes-uninitialized]
> > if (!longmode) {
> > ^~~~~~~~~
> > arch/x86/kvm/hyperv.c:1632:55: note: uninitialized use occurs here
> > trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
> > ^~~~~
> > arch/x86/kvm/hyperv.c:1610:2: note: remove the 'if' if its condition is always true
> > if (!longmode) {
> > ^~~~~~~~~~~~~~~
> > arch/x86/kvm/hyperv.c:1595:18: note: initialize the variable 'ingpa' to silence this warning
> > u64 param, ingpa, outgpa, ret = HV_STATUS_SUCCESS;
> > ^
> > = 0
> > arch/x86/kvm/hyperv.c:1610:6: error: variable 'outgpa' is used uninitialized whenever 'if' condition is false [-Werror,-Wsometimes-uninitialized]
> > arch/x86/kvm/hyperv.c:1610:6: error: variable 'param' is used uninitialized whenever 'if' condition is false [-Werror,-Wsometimes-uninitialized]
> >
> > Since that combination is not supported anyway, change the condition
> > to tell the compiler how the code is actually executed.
> >
> > Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
> > ---
> > v2: make the change inside of is_64_bit_mode().
> > ---
> > arch/x86/kvm/hyperv.c | 6 ++----
> > arch/x86/kvm/x86.h | 4 ++++
> > 2 files changed, 6 insertions(+), 4 deletions(-)
>
> Reviewed-by: Roman Kagan <rkagan@xxxxxxxxxxxxx>
>
> However I still think the log message could state it more explicitly
> that it was the compiler's fault, and the patch is a workaround for it.
>
> Otherwise later on someone may decide to restore the similarity of
> is_64_bit_mode() to other inlines in this file, and will be extremely
> unlikely to test clang 32-bit build...
Fair enough. I've reworded the changelog, as well as the patch to
document this now, in a way that should make it harder to introduce
the warning again by accident. Unfortunately, that #ifdef check
cannot be turned into an if(IS_ENABLED()) because kvm_r8_read()
is not defined on i386.
Note that the 0-day bot now tests with clang as well, so you would
definitely hear about a warning appearing.
v3 coming.
Arnd
