On Tue, 24 Feb 2026, Peter Maydell wrote:
On Tue, 24 Feb 2026 at 17:45, BALATON Zoltan <[email protected]> wrote:
On Tue, 24 Feb 2026, Peter Xu wrote:
On Thu, Feb 05, 2026 at 02:13:18AM +0100, BALATON Zoltan wrote:
v4:
- separate patch converting Sun machines from memory_region_init_ram_nomigrate
- split helper to init ram into two functions: setup and error_propagate
- also use memory_region_init_io in memory_region_init_ram_device_ptr
v3:
- rebased on master after some patches were merged
- drop some more line from memory-region-housekeeping.cocci
- added comment to explain what factored out helper does
- some more clean ups included
BALATON Zoltan (8):
hw/display/{cg3,tcx}: Do not use memory_region_init_rom_nomigrate()
memory: Remove memory_region_init_rom_nomigrate()
sun4m,sun4u,tcx: Do not use memory_region_init_ram_nomigrate()
memory: Remove memory_region_init_ram_nomigrate()
Could you help explain why we need to remove the above two small helpers?
Ideally we should remove all of the nomigrate variants as there are only
about 3 uses left, everything else is already moved to the init_{ram,rom}
variants that register the memory region for migration. The is probably
one legitimate use that is served by memory_region_init_flags_nomigrate
and other 3 uses are the xtfpga which I think is just old code and does
not support migration so it will be converted in the second series to
memory_region_new as the maintainers did not say anyting to that so far.
Then there's one in hw/display/vga but that's gated behind a compatibility
switch that's only set in 2.12 (global-vmstate=true) and earlier. This is
not active any more but cannot yet be removed as there are eariler compat
props still need cleaning up but the last visible machine version is 5.1
now so this is long deprecated and not used. And there are these Sun
machines that I've converted in the first version but Mark said he'd like
to keep the current behaviour until it's the last user left. See:
https://patchew.org/QEMU/[email protected]/b530cf54609b853c6b34306fd4aa3ce2f8087d05.1769703287.git.bala...@eik.bme.hu/
Not saying that we can't remove them, but I don't yet see the point of
open-code those readonly=true setup either, or any problem with having the
smaller helpers when ram_flags=0. Small helpers sometimes help readability.
Some of them are going backwards againist what the cocci scripts were
suggesting previously, like:
- memory_region_init_ram_nomigrate(E1, E2, E3, E4, E5);
+ memory_region_init_rom_nomigrate(E1, E2, E3, E4, E5);
... WHEN != E1
- memory_region_set_readonly(E1, true);
So you suggest open-code instead. Why?
This is temporary until the usage in vga is gone with hw_compat_2_12 then
the Sun machines will be the last users so they can also be converted to
init_{ram,rom} (unless Mark changes his mind and agrees changing these
now as was in the first version).
I think we should take the migration-compat break in the Sun machines
for the sake of cleaning up those callsites now. Sun isn't quite the
last thing standing using these functions, but it almost is, and as
you say the xtfpga can go too.
Mark, would it be OK to convert the Sun machines to use
memory_region_init_rom now so we can remove the nomigrate variant more
easily? The xtfpga can be converted now and the usage in vga is only there
because the 2.12 hw_compat is not yet removed but is not anymore used and
should be gone eventually so at the latest then these usages will be the
last one and would be converted. If you agree to not wait until then, I
can send another version of this clean up series with the first version of
the patch that converts the Sun display devices to memory_region_init_rom.
Regards,
BALATON Zoltan
