xen/pci: support multi-segment systems

Now that the hypercall interface changes are in -unstable, make the
kernel side code not ignore the segment (aka domain) number anymore
(which results in pretty odd behavior on such systems). Rather, if
only the old interfaces are available, don't call them for devices on
non-zero segments at all.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
[v1: Edited git description]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>

1 files changed, 84 insertions, 10 deletions

diff --git a/drivers/xen/pci.c b/drivers/xen/pci.c
index 02c402b..6605707 100644
--- a/drivers/xen/pci.c
+++ b/drivers/xen/pci.c
@@ -18,6 +18,7 @@
  */
 
 #include <linux/pci.h>
+#include <linux/acpi.h>
 #include <xen/xen.h>
 #include <xen/interface/physdev.h>
 #include <xen/interface/xen.h>
@@ -26,26 +27,85 @@
 #include <asm/xen/hypercall.h>
 #include "../pci/pci.h"
 
+static bool __read_mostly pci_seg_supported = true;
+
 static int xen_add_device(struct device *dev)
 {
 	int r;
 	struct pci_dev *pci_dev = to_pci_dev(dev);
+#ifdef CONFIG_PCI_IOV
+	struct pci_dev *physfn = pci_dev->physfn;
+#endif
+
+	if (pci_seg_supported) {
+		struct physdev_pci_device_add add = {
+			.seg = pci_domain_nr(pci_dev->bus),
+			.bus = pci_dev->bus->number,
+			.devfn = pci_dev->devfn
+		};
+#ifdef CONFIG_ACPI
+		acpi_handle handle;
+#endif
 
 #ifdef CONFIG_PCI_IOV
-	if (pci_dev->is_virtfn) {
+		if (pci_dev->is_virtfn) {
+			add.flags = XEN_PCI_DEV_VIRTFN;
+			add.physfn.bus = physfn->bus->number;
+			add.physfn.devfn = physfn->devfn;
+		} else
+#endif
+		if (pci_ari_enabled(pci_dev->bus) && PCI_SLOT(pci_dev->devfn))
+			add.flags = XEN_PCI_DEV_EXTFN;
+
+#ifdef CONFIG_ACPI
+		handle = DEVICE_ACPI_HANDLE(&pci_dev->dev);
+		if (!handle)
+			handle = DEVICE_ACPI_HANDLE(pci_dev->bus->bridge);
+#ifdef CONFIG_PCI_IOV
+		if (!handle && pci_dev->is_virtfn)
+			handle = DEVICE_ACPI_HANDLE(physfn->bus->bridge);
+#endif
+		if (handle) {
+			acpi_status status;
+
+			do {
+				unsigned long long pxm;
+
+				status = acpi_evaluate_integer(handle, "_PXM",
+							       NULL, &pxm);
+				if (ACPI_SUCCESS(status)) {
+					add.optarr[0] = pxm;
+					add.flags |= XEN_PCI_DEV_PXM;
+					break;
+				}
+				status = acpi_get_parent(handle, &handle);
+			} while (ACPI_SUCCESS(status));
+		}
+#endif /* CONFIG_ACPI */
+
+		r = HYPERVISOR_physdev_op(PHYSDEVOP_pci_device_add, &add);
+		if (r != -ENOSYS)
+			return r;
+		pci_seg_supported = false;
+	}
+
+	if (pci_domain_nr(pci_dev->bus))
+		r = -ENOSYS;
+#ifdef CONFIG_PCI_IOV
+	else if (pci_dev->is_virtfn) {
 		struct physdev_manage_pci_ext manage_pci_ext = {
 			.bus		= pci_dev->bus->number,
 			.devfn		= pci_dev->devfn,
 			.is_virtfn 	= 1,
-			.physfn.bus	= pci_dev->physfn->bus->number,
-			.physfn.devfn	= pci_dev->physfn->devfn,
+			.physfn.bus	= physfn->bus->number,
+			.physfn.devfn	= physfn->devfn,
 		};
 
 		r = HYPERVISOR_physdev_op(PHYSDEVOP_manage_pci_add_ext,
 			&manage_pci_ext);
-	} else
+	}
 #endif
-	if (pci_ari_enabled(pci_dev->bus) && PCI_SLOT(pci_dev->devfn)) {
+	else if (pci_ari_enabled(pci_dev->bus) && PCI_SLOT(pci_dev->devfn)) {
 		struct physdev_manage_pci_ext manage_pci_ext = {
 			.bus		= pci_dev->bus->number,
 			.devfn		= pci_dev->devfn,
@@ -71,13 +131,27 @@ static int xen_remove_device(struct device *dev)
 {
 	int r;
 	struct pci_dev *pci_dev = to_pci_dev(dev);
-	struct physdev_manage_pci manage_pci;
 
-	manage_pci.bus = pci_dev->bus->number;
-	manage_pci.devfn = pci_dev->devfn;
+	if (pci_seg_supported) {
+		struct physdev_pci_device device = {
+			.seg = pci_domain_nr(pci_dev->bus),
+			.bus = pci_dev->bus->number,
+			.devfn = pci_dev->devfn
+		};
 
-	r = HYPERVISOR_physdev_op(PHYSDEVOP_manage_pci_remove,
-		&manage_pci);
+		r = HYPERVISOR_physdev_op(PHYSDEVOP_pci_device_remove,
+					  &device);
+	} else if (pci_domain_nr(pci_dev->bus))
+		r = -ENOSYS;
+	else {
+		struct physdev_manage_pci manage_pci = {
+			.bus = pci_dev->bus->number,
+			.devfn = pci_dev->devfn
+		};
+
+		r = HYPERVISOR_physdev_op(PHYSDEVOP_manage_pci_remove,
+					  &manage_pci);
+	}
 
 	return r;
 }