[libvirt] [PATCH v4 19/25] [ACKED] qemu: only force an available legacy-PCI slot on domains with pci-root

[Laine Stump]

Andrea had the right idea when he disabled the "reserve an extra
unused slot" bit for aarch64/virt. For *any* PCI Express-based
machine, it is pointless since 1) an extra legacy-PCI slot can't be
used for hotplug, since hotplug into legacy PCI slots doesn't work on
PCI Express machinetypes, and 2) even for "coldplug" expansion,
everybody will want to expand using Express controllers, not legacy
PCI.

This patch eliminates the extra slot reserve unless the system has a
pci-root (i.e. legacy PCI)
---
 src/qemu/qemu_domain_address.c | 44 ++++++++++++++++++++++++++----------------
 1 file changed, 27 insertions(+), 17 deletions(-)

diff --git a/src/qemu/qemu_domain_address.c b/src/qemu/qemu_domain_address.c
index 5262ad3..4e4e0d6 100644
--- a/src/qemu/qemu_domain_address.c
+++ b/src/qemu/qemu_domain_address.c
@@ -1815,8 +1815,6 @@ qemuDomainAssignPCIAddresses(virDomainDefPtr def,
          */
         virDomainDeviceInfo info;
 
-        info.pciConnectFlags = VIR_PCI_CONNECT_TYPE_PCI_DEVICE;
-
         /* 1st pass to figure out how many PCI bridges we need */
         if (!(addrs = qemuDomainPCIAddressSetCreate(def, nbuses, true)))
             goto cleanup;
@@ -1825,23 +1823,35 @@ qemuDomainAssignPCIAddresses(virDomainDefPtr def,
                                                      addrs) < 0)
             goto cleanup;
 
-        for (i = 0; i < addrs->nbuses; i++) {
-            if (!qemuDomainPCIBusFullyReserved(&addrs->buses[i]))
-                buses_reserved = false;
-        }
-
-        /* Reserve 1 extra slot for a (potential) bridge only if buses
-         * are not fully reserved yet.
+        /* For domains that have pci-root, reserve 1 extra slot for a
+         * (potential) bridge (for future expansion) only if buses are
+         * not fully reserved yet (if all buses are fully reserved
+         * with manually/previously assigned addresses, any attempt to
+         * reserve an extra slot would fail anyway. But if all buses
+         * are *not* fully reserved, this extra reservation might push
+         * the config to add a new pci-bridge to plug into the final
+         * available slot, thus preserving the ability to expand)
          *
-         * We don't reserve the extra slot for aarch64 mach-virt guests
-         * either because we want to be able to have pure virtio-mmio
-         * guests, and reserving this slot would force us to add at least
-         * a dmi-to-pci-bridge to an otherwise PCI-free topology
+         * We only do this for those domains that have pci-root, since
+         * those with pcie-root will usually want to expand using PCIe
+         * controllers, which we will do after assigning addresses for
+         * all *actual* devices.
          */
-        if (!buses_reserved &&
-            !qemuDomainMachineIsVirt(def) &&
-            qemuDomainPCIAddressReserveNextSlot(addrs, &info) < 0)
-            goto cleanup;
+
+        if (qemuDomainMachineHasPCIRoot(def)) {
+            info.pciConnectFlags = (VIR_PCI_CONNECT_HOTPLUGGABLE |
+                                    VIR_PCI_CONNECT_TYPE_PCI_DEVICE);
+
+            for (i = 0; i < addrs->nbuses; i++) {
+                if (!qemuDomainPCIBusFullyReserved(&addrs->buses[i])) {
+                    buses_reserved = false;
+                    break;
+                }
+            }
+            if (!buses_reserved &&
+                qemuDomainPCIAddressReserveNextSlot(addrs, &info) < 0)
+                goto cleanup;
+        }
 
         if (qemuDomainAssignDevicePCISlots(def, qemuCaps, addrs) < 0)
             goto cleanup;
-- 
2.7.4