On 20 March 2018 at 11:36, Shannon Zhao <zhaoshengl...@huawei.com> wrote: > > > On 2018/3/20 19:22, Peter Maydell wrote: >> On 20 March 2018 at 07:26, Shannon Zhao <zhaoshengl...@huawei.com> wrote: >>> While we skip the GIC_INTERNAL irqs, we don't change the register offset >>> accordingly. This will overlap the GICR registers value and leave the >>> last GIC_INTERNAL irq's registers out of update. >>> >>> Fix this by skipping the registers banked by GICR. >>> >> >> I'm still not entirely sure what the underlying problem >> you're trying to fix is... >> >> Do we fail to correctly migrate a VM without this change? >> Does the code work on some host CPU/GIC implementations but >> not others? Is this just improving efficiency by avoiding >> doing some unnecessary work? >> > When we reboot a VM and before entering uefi or guest kernel, we expect > all these registers staying at the initial state. But currently these > registers of the last 32 irqs are not reset. For example, the PRIORITY > of irq from 32 to 255 is 0 but the PRIORITY of irq from 256 to 287 is > 0xa0(Linux kernel set the PRIORITY to 0xa0 by default). > > When migrating a VM, since we don't save and restore the registers of > the last 32 irq, so the PRIORITY is 0 while we expecting 0xa0. > And also it will overlap the PRIORITY of SGIs and PPIs. > > We don't fail to migrate a vm since currently we don't use the last 32 > irqs in virt machine. But the bug is still there.
Oh, I see, the number of registers we transfer is accounting for the first N registers in the bank not being used, but the first register offset to transfer wasn't. Can you still successfully migrate a VM from a QEMU version without this bugfix to one with the bugfix ? thanks -- PMM