ACPI / hotplug / PCI: Runtime resume bridges before bus rescans

If a PCI bridge (or PCIe port) that is runtime-suspended gets an ACPI
hotplug notification, such as a bus check, it has to be resumed before
re-scanning the devices below it, or those devices will not be
accessible and will be treated as hot-removed.

Make that happen and let the bridge suspend again after the bus below it
has been re-scanned.

This is a replacement for commit 16468c78 ("ACPI / hotplug / PCI:
Runtime resume bridge before rescan") that has been reverted, because it
introduced a system resume regression (due to missing bridge->pci_dev
checks that are necessary in case the notification is targeted at the
host bridge) and it is necessary for the code added by commit
006d44e4 ("PCI: Add runtime PM support for PCIe ports") to work as
expected.
Tested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/pci/hotplug/acpiphp_glue.c

@@ -675,6 +675,9 @@ static void acpiphp_check_bridge(struct acpiphp_bridge *bridge)
 	if (bridge->is_going_away)
 		return;
 
+	if (bridge->pci_dev)
+		pm_runtime_get_sync(&bridge->pci_dev->dev);
+
 	list_for_each_entry(slot, &bridge->slots, node) {
 		struct pci_bus *bus = slot->bus;
 		struct pci_dev *dev, *tmp;
@@ -694,6 +697,9 @@ static void acpiphp_check_bridge(struct acpiphp_bridge *bridge)
 			disable_slot(slot);
 		}
 	}
+
+	if (bridge->pci_dev)
+		pm_runtime_put(&bridge->pci_dev->dev);
 }
 
 /*