hv: vmsi: add vmsix on msi emulation support

Some passthrough devices require multiple MSI vectors, but don't
support MSI-X. In meanwhile, Linux kernel doesn't support continuous
vector allocation.
On native platform, this issue can be mitigated by IOMMU via interrupt
remapping. However, on ACRN, there is no vIOMMU.
vMSI-X on MSI emulation is one solution to mitigate this problem on ACRN.

This patch adds MSI-X emulation on MSI capability.
For the device needs to do MSI-X emulation, HV will hide MSI capability
and present MSI-X capability to guest.

The guest driver may need to modify to reqeust MSI-X vector.
For example:
        ret = pci_alloc_irq_vectors(pdev, 1, STMMAC_MSI_VEC_MAX,
-                                   PCI_IRQ_MSI);
+                                   PCI_IRQ_MSI | PCI_IRQ_MSIX);

To enable MSI-X emulation, the device should:
- 1. The device should be in vmsix_on_msi_devs array.
- 2. Support MSI, but don't support MSI-X.
- 3. MSI capability should support per-vector mask.
- 4. The device should have an unused BAR.
- 5. The device driver should not rely on PBA for functionality.

Tracked-On: #4831
Signed-off-by: Binbin Wu <binbin.wu@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>

hypervisor/include/hw/pci.h

@@ -228,6 +228,9 @@ struct pci_pdev {
 
 	/* IOMMU responsible for DMA and Interrupt Remapping for this device */
 	uint32_t drhd_index;
+	/* Used for vMSI-x on MSI emulation */
+	uint16_t irte_start;
+	uint16_t irte_count;
 
 	/* The bar info of the physical PCI device. */
 	uint32_t nr_bars; /* 6 for normal device, 2 for bridge, 1 for cardbus */
@@ -359,4 +362,5 @@ bool is_plat_hidden_pdev(union pci_bdf bdf);
 bool pdev_need_bar_restore(const struct pci_pdev *pdev);
 void pdev_restore_bar(const struct pci_pdev *pdev);
 void pci_switch_to_mmio_cfg_ops(void);
+void reserve_vmsix_on_msi_irtes(struct pci_pdev *pdev);
 #endif /* PCI_H_ */