qemu aarch64 has a little bug when using msi

[liulog]

Environment

• root0: one virtio-blk-pci
• root1: one virtio-blk-pci

Phenomenon

The non-root can't work well when using one blk-pci device.

But when I configure it use wire-interrupt, it works well.

Conclusion

So I think message-interrupt in current hvisor has a little bug.
@ZhongkaiXu

[ZhongkaiXu]

Thank you for reporting this issue.
The problem occurs because Linux uses logical ICIDs when configuring the ITS, rather than relying on the cpuid from the device tree.
Even though we have already performed the remap in the MAPC phase, ignoring this may still allow later updates through a MOVI command.
I’ll share a suggested code change below and submit a PR soon, linking it to your issue.

// gits.rs/impl Cmdq/analyze_cmd

    fn analyze_cmd(&self, value: [u64; 4]) -> [u64; 4] {
        let code = (value[0] & 0xff) as usize;
        let mut new_cmd = value.clone();
        let binding = this_zone();
        let zone = binding.read();
        let cpuset_bitmap = zone.cpu_set.bitmap;
        match code {
            // ...
            // ----add movi-cmd remap here----
            0x01 => {
                let vicid = value[2] & 0xffff;
                let icid = vicid_to_icid(vicid, cpuset_bitmap)
                    .expect("vicid to icid failed, maybe logical_id out of range");
                new_cmd[2] &= !0xffffu64;
                new_cmd[2] |= icid & 0xffff;
                info!("MOVI, for Device {:#x}, new vicid({}) -> icid({})", value[0]>>32, vicid, icid);
            }
            // -----------code end--------------
            _ => {
                info!("other cmd, code: 0x{:x}", code);
            }
        }

        new_cmd
    }