fix(loongarch64): harden virtio IRQ and zone startup

[li041]

Summary

• model LoongArch virtio interrupt lines with explicit assertion state
• harden IVC MMIO/IRQ initialization and isolate the canonical virtio sources
• return a failure status when the zone-start ioctl fails

Validation

• loaded hvisor.ko and started linux1 on the LS3A6000 board
• verified /dev/pts/0 login, uname, and two guest CPUs
• verified zone-start failures are visible to the calling script

[li041]

@enkerewpo Thanks for your review. I have addressed the comments and updated the PR accordingly. Could you please take another look when you have time?

[agicy]

I recently hit a console problem in my multi-zone setup: screen zone1 -> exit -> screen zone2 -> back to zone1, and the console is sometimes completely frozen after re-attach. So I'm glad to see this PR touching virtio-console - I did a careful review and left inline comments. Summary of what I found:

1. TX backpressure is still missing. On EAGAIN (pty full), the descriptor is still completed via update_used_ring() and freed, so the guest is told its output was transmitted. A flooding guest never throttles and all output is silently dropped while the pty stays full - this is exactly the "console frozen after re-attach" symptom I observed. The RX changes in this PR don't address it.
2. The RX rearm logic introduces a permanent stall. rxq_notify_handler calls disable_notify(), which sets VRING_USED_F_NO_NOTIFY. But the event handler only re-arms when the queue is non-empty. If it exits with an empty queue while input is still pending in the pty, the EPOLLONESHOT event stays disarmed forever and the guest's next kick never comes -> RX is permanently dead.
3. The descriptor-error path never re-arms the event, so one transient process_descriptor_chain() failure also kills console RX permanently.

Since these console changes may carry a regression and the actual backpressure fix is still missing, I'd suggest extracting the virtio-console changes from this PR and landing them as a separate PR, so the console fix can be completed properly (RX improvements + rearm fix + TX backpressure via EPOLLOUT) and merged independently of the LoongArch/IVC work.

[li041]

Since these console changes may carry a regression and the actual backpressure fix is still missing, I'd suggest extracting the virtio-console changes from this PR and landing them as a separate PR, so the console fix can be completed properly (RX improvements + rearm fix + TX backpressure via EPOLLOUT) and merged independently of the LoongArch/IVC work.

Thanks. This PR now only includes fix on la irq and ivc. A Latter PR will focus on virtio console RX/TX. Maybe we can discuss in that PR carefully.

[enkerewpo]

Thanks for the fix — the level-triggered interrupt model and the interrupt_status &= !value → & ~ack correction are both clear improvements (the old logical-not was a real bug).

Two blocking issues before this can land, both about how the new deassert path is numbered, plus a leak on the ivc error path. Details inline.

[enkerewpo]

LGTM