Re: [RFC PATCH] target/ppc/mmu: Silent maybe-uninitialized error in ppc_hash64_xlate()

[Thomas Huth]

On 23/02/2024 09.32, Philippe Mathieu-Daudé wrote:
> Initialize apshift to avoid a maybe-uninitialized error:
>
>    C compiler for the host machine: cc -m64 -mbig-endian (gcc 13.2.0 "cc (Debian 
> 13.2.0-10) 13.2.0")
>    C linker for the host machine: cc -m64 -mbig-endian ld.bfd 2.41.90.20240115
>    Host machine cpu family: ppc64
>    Host machine cpu: ppc64
>    ...
>    target/ppc/mmu-hash64.c: In function 'ppc_hash64_xlate':
>    target/ppc/mmu-hash64.c:1154:15: error: 'apshift' may be used uninitialized 
> [-Werror=maybe-uninitialized]
>     1154 |     *raddrp = deposit64(pte.pte1 & HPTE64_R_RPN, 0, apshift, eaddr);
>          |               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>    target/ppc/mmu-hash64.c:947:14: note: 'apshift' was declared here
>      947 |     unsigned apshift;
>          |              ^~~~~~~
>
> The call chain is:
>
>    ppc_hash64_xlate -> ppc_hash64_htab_lookup -> ppc_hash64_pteg_search
>
> ppc_hash64_pteg_search() either sets *pshift or returns -1,
>
> ppc_hash64_htab_lookup() returns if ppc_hash64_pteg_search()
> returned -1:
>
>    1068:    ptex = ppc_hash64_htab_lookup(cpu, slb, eaddr, &pte, &apshift);
>    1069:    if (ptex == -1) {
>    1070:        if (!guest_visible) {
>    1071:            return false;
>    1072:        }
>                 ...
>    1087:        return false;
>
> So IIUC this "uninitialized use" can not happens.
>
> Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> ---
> I had this in an old branch (2 months old) I just rebased,
> and don't get why nobody else got this error yet.

That's weird, indeed. Did you maybe compile without optimizations when you 
hit the error?

 Thomas