On 12/02/2021 13.05, Peter Maydell wrote:
On Fri, 12 Feb 2021 at 11:45, Thomas Huth <th...@redhat.com> wrote:
I guess Cornelia simply meant a:
asm volatile("nop":::"memory");
That will force the compiler not to reorder, but it doesn't
seem to me that it would really force the memory accesses to
appear to the other host CPU thread that's running the device
emulation. So either it's insufficient in the s390 memory model,
or our s390 emulation isn't being sufficiently rigorous about
implementing the memory model...
AFAIK s390x has a very strict memory ordering model... so maybe this is a
problem with emulating a strict CPU on a less strict host CPU, indeed? ...
CC:-ing Christian and Richard, maybe they have some additional ideas here...
Thomas
