P J P <ppan...@redhat.com> 于2020年6月25日周四 上午3:00写道： > > From: Prasad J Pandit <p...@fedoraproject.org> > > When registering a MemoryRegionOps object, assert that its > read/write callback methods are defined. This avoids potential > guest crash via a NULL pointer dereference. > > Suggested-by: Peter Maydell <peter.mayd...@linaro.org> > Signed-off-by: Prasad J Pandit <p...@fedoraproject.org> > --- > memory.c | 10 +++++++++- > 1 file changed, 9 insertions(+), 1 deletion(-) > > Update v2: assert read/write_with_attrs methods > -> https://lists.gnu.org/archive/html/qemu-devel/2020-06/msg05717.html > > diff --git a/memory.c b/memory.c > index 2f15a4b250..3d635a1bca 100644 > --- a/memory.c > +++ b/memory.c > @@ -1496,7 +1496,13 @@ void memory_region_init_io(MemoryRegion *mr, > uint64_t size) > { > memory_region_init(mr, owner, name, size); > - mr->ops = ops ? ops : &unassigned_mem_ops; > + if (ops) { > + assert(ops->read || ops->read_with_attrs); > + assert(ops->write || ops->write_with_attrs); > + mr->ops = ops; > + } else { > + mr->ops = &unassigned_mem_ops; > + } > mr->opaque = opaque; > mr->terminates = true; > } > @@ -1674,6 +1680,8 @@ void > memory_region_init_rom_device_nomigrate(MemoryRegion *mr, > { > Error *err = NULL; > assert(ops); > + assert(ops->read || ops->read_with_attrs);
Though here is not 100% right as the ROM device can has no read callback. However the device can change between ROMD mode or MMIO mode. So I think it's ok the developer can provide a null read callback even if they don't need it. Reviewed-by: Li Qiang <liq...@gmail.com> > + assert(ops->write || ops->write_with_attrs); > memory_region_init(mr, owner, name, size); > mr->ops = ops; > mr->opaque = opaque; > -- > 2.26.2 >
