[Qemu-stable] get_pci_irq_state causes inconsistent pci_irq_state

[Xusheng Chen]

Dear qemu developers, 

I am interested in qemu and found one bug that crash colo-fault-tolerance because of inconsistent PCI state. 

It is confirmed by the colo developers and they said it is likely to be a generic bug that will also break migration. 

When the receiver of a device state calls get_pci_irq_state(hw/pci/pci.c L506), it only set the irq_state of the device but not that of the PCI bus. As a result, when the sync is doing very frequently, there may be inconsistencies between the irq_state of the PCI bus and the devices, causing the assertion(L311) failed and the whole system halted.

One possible patch is as follows:

--- a/hw/pci/pci.c

+++ b/hw/pci/pci.c

@@ -520,6 +520,10 @@ static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,

 

     for (i = 0; i < PCI_NUM_PINS; ++i) {

         pci_set_irq_state(s, i, irq_state[i]);

+        if (irq_state[i] != 0){

+            s->bus->set_irq(s->bus->irq_opaque, PCI_NUM_PINS-1-i , 1);

+            s->bus->irq_count[PCI_NUM_PINS-1-i] = 1; 

+        }

     }

Thanks and regards,

Xusheng Chen