(+Markus I forgot to Cc)
On 9/2/24 17:06, Peter Maydell wrote:
On Fri, 9 Feb 2024 at 15:01, Philippe Mathieu-Daudé <[email protected]> wrote:
So far there is only one system I/O and one system
memory per machine.
Signed-off-by: Philippe Mathieu-Daudé <[email protected]>
---
system/physmem.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)
diff --git a/system/physmem.c b/system/physmem.c
index 5e66d9ae36..50947a374e 100644
--- a/system/physmem.c
+++ b/system/physmem.c
@@ -2554,12 +2554,13 @@ static void memory_map_init(void)
{
system_memory = g_malloc(sizeof(*system_memory));
- memory_region_init(system_memory, NULL, "system", UINT64_MAX);
+ memory_region_init(system_memory, OBJECT(current_machine),
+ "system", UINT64_MAX);
address_space_init(&address_space_memory, system_memory, "memory");
system_io = g_malloc(sizeof(*system_io));
- memory_region_init_io(system_io, NULL, &unassigned_io_ops, NULL, "io",
- 65536);
+ memory_region_init_io(system_io, OBJECT(current_machine),
+ &unassigned_io_ops, NULL, "io", 65535);
address_space_init(&address_space_io, system_io, "I/O");
}
What's the intention in doing this? What does it change?
We want to remove access to pre-QOM and possibly hotplug QOM paths
from external API (CLI & QMP so far).
When the parent object is obvious and missing we simply have to
explicit it.
It seems to be OK to pass a non-Device owner in for
memory_region_init() (whereas it is *not* OK to do that
for memory_region_init_ram()), but this seems to be
getting a bit tricky.
Yes, memory_region_init_ram() is problematic; I'm hardly trying
to ignore it at this point.
