On Wed, 26 Sep 2018 11:46:22 +0200 Thomas Huth <[email protected]> wrote:
> On 2018-09-26 11:42, Cornelia Huck wrote: > > On Wed, 26 Sep 2018 09:38:45 +0200 > > Thomas Huth <[email protected]> wrote: > > > >> The IplParameterBlock and QemuIplParameters structures are declared > >> with QEMU_PACKED, so the compiler assumes that the structures do not > >> need to be aligned in memory. Since the are listed after a "bool" > >> within the S390IPLState, the IplParameterBlock and QemuIplParameters > >> are also indeed mis-aligned in memory. This causes problems on Sparc > >> during migration, since we use VMSTATE_UINT16 in vmstate_iplb to access > >> the devno member for example, and the corresponding migration functions > >> (like qemu_get_be16s) then try to access a 16-bit value from a mis- > >> aligned memory address. > >> The easiest solution to fix this problem is to move the packed structures > >> to the beginning of the S390IPLState. Also add some additional comments > >> here to prevent that this problem will be introduced again in the future. > >> > >> Signed-off-by: Thomas Huth <[email protected]> > >> --- > >> hw/s390x/ipl.h | 5 +++-- > >> 1 file changed, 3 insertions(+), 2 deletions(-) > >> > >> diff --git a/hw/s390x/ipl.h b/hw/s390x/ipl.h > >> index 4e87b89..b3a07a1 100644 > >> --- a/hw/s390x/ipl.h > >> +++ b/hw/s390x/ipl.h > >> @@ -132,15 +132,15 @@ typedef struct QemuIplParameters QemuIplParameters; > >> struct S390IPLState { > >> /*< private >*/ > >> DeviceState parent_obj; > >> + IplParameterBlock iplb; > >> + QemuIplParameters qipl; > > > > Hm... this is not quite the beginning of the structure; what am I > > missing? > > DeviceState of course has to stay first for QOM reasons. But since it is > a non-packed struct, we can be sure that it will be padded to the > correct alignment at the end. If not, the QEMU_BUILD_BUG_MSG in this > patch will tell us. What about adding that explanation to the commit message? [I can do that.]
