On 2018-12-10 14:58, Peter Maydell wrote:
> Taking the address of a field in a packed struct is a bad idea, because
> it might not be actually aligned enough for that pointer type (and
> thus cause a crash on dereference on some host architectures). Newer
> versions of clang warn about this.
>
> Avoid the problem by using local copies of the PMCW and SCSW
> struct fields in copy_schib_from_guest() and copy_schib_to_guest().
>
> Signed-off-by: Peter Maydell <peter.mayd...@linaro.org>
> ---
> This seemed like a not totally ugly and reasonably localised fix
> that satisfies clang. Oddly, this makes the generated object file
> 15K smaller (421K vs 406K), so it might even be better code...
>
> hw/s390x/css.c | 20 ++++++++++++++++----
> 1 file changed, 16 insertions(+), 4 deletions(-)
>
> diff --git a/hw/s390x/css.c b/hw/s390x/css.c
> index 04ec5cc9705..ef07691e36b 100644
> --- a/hw/s390x/css.c
> +++ b/hw/s390x/css.c
> @@ -1290,9 +1290,15 @@ void copy_scsw_to_guest(SCSW *dest, const SCSW *src)
> static void copy_schib_to_guest(SCHIB *dest, const SCHIB *src)
> {
> int i;
> + PMCW srcpmcw, destpmcw;
> + SCSW srcscsw, destscsw;
>
> - copy_pmcw_to_guest(&dest->pmcw, &src->pmcw);
> - copy_scsw_to_guest(&dest->scsw, &src->scsw);
> + srcpmcw = src->pmcw;
> + copy_pmcw_to_guest(&destpmcw, &srcpmcw);
> + dest->pmcw = destpmcw;
> + srcscsw = src->scsw;
> + copy_scsw_to_guest(&destscsw, &srcscsw);
> + dest->scsw = destscsw;
> dest->mba = cpu_to_be64(src->mba);
> for (i = 0; i < ARRAY_SIZE(dest->mda); i++) {
> dest->mda[i] = src->mda[i];
> @@ -1339,9 +1345,15 @@ static void copy_scsw_from_guest(SCSW *dest, const
> SCSW *src)
> static void copy_schib_from_guest(SCHIB *dest, const SCHIB *src)
> {
> int i;
> + PMCW srcpmcw, destpmcw;
> + SCSW srcscsw, destscsw;
>
> - copy_pmcw_from_guest(&dest->pmcw, &src->pmcw);
> - copy_scsw_from_guest(&dest->scsw, &src->scsw);
> + srcpmcw = src->pmcw;
> + copy_pmcw_from_guest(&destpmcw, &srcpmcw);
> + dest->pmcw = destpmcw;
> + srcscsw = src->scsw;
> + copy_scsw_from_guest(&destscsw, &srcscsw);
> + dest->scsw = destscsw;
> dest->mba = be64_to_cpu(src->mba);
> for (i = 0; i < ARRAY_SIZE(dest->mda); i++) {
> dest->mda[i] = src->mda[i];
>
May I suggest to add a comment to the code here a la:
/* Use a local copy to avoid unaligned access to packed structs */
or something similar? Otherwise, I'm pretty sure somebody will revert
this in a couple of years because they thinks the local copy is not
really necessary here...
Thomas
