On 14/1/22 14:36, Peter Maydell wrote:
On Wed, 12 Jan 2022 at 22:02, Bernhard Beschow <shen...@gmail.com> wrote:Now that piix4_set_irq's opaque parameter references own PIIX4State, piix4_dev becomes redundant and pci_irq_levels can be moved into PIIX4State. Signed-off-by: Bernhard Beschow <shen...@gmail.com> --- hw/isa/piix4.c | 22 +++++++++------------- include/hw/southbridge/piix.h | 2 -- 2 files changed, 9 insertions(+), 15 deletions(-) diff --git a/hw/isa/piix4.c b/hw/isa/piix4.c index a31e9714cf..964e09cf7f 100644 --- a/hw/isa/piix4.c +++ b/hw/isa/piix4.c @@ -39,14 +39,14 @@ #include "sysemu/runstate.h" #include "qom/object.h" -PCIDevice *piix4_dev; - struct PIIX4State { PCIDevice dev; qemu_irq cpu_intr; qemu_irq *isa; qemu_irq i8259[ISA_NUM_IRQS]; + int pci_irq_levels[PIIX_NUM_PIRQS]; +I wondered how we were migrating this state, and the answer seems to be that we aren't (and weren't before, when it was a global variable, so this is a pre-existing bug).
Indeed the migrated VM starts with PCI IRQ levels zeroed.
Does the malta platform support migration save/load?
Maybe a "best effort" support, but not versioned machines.
We should probably add this field to the vmstate struct (which will be a migration compatibility break, which is OK as the malta board isn't versioned.)
Yeah good catch. Bernhard, do you mind adding it? Regards, Phil.
