Konrad Rzeszutek Wilk
2013-Nov-08 17:38 UTC
[PATCH 1/4] xen/mcfg: Call PHYSDEVOP_pci_mmcfg_reserved for MCFG areas.
The PCI MMCONFIG area is usually reserved via the E820 so the Xen hypervisor
is aware of these regions. But they can also be enumerated in the ACPI
DSDT which means the hypervisor won''t know of them until the initial
domain informs it of via PHYSDEVOP_pci_mmcfg_reserved.
This is what this patch does for all of the MCFG regions that the
initial domain is aware of (E820 enumerated and ACPI).
Reported-by: Santosh Jodh <Santosh.Jodh@citrix.com>
CC: Jan Beulich <JBeulich@suse.com>
CC: Boris Ostrovsky <boris.ostrovsky@oracle.com>
CC: David Vrabel <david.vrabel@citrix.com>
CC: Mukesh Rathor <mukesh.rathor@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
[v1: Redid it a bit]
[v2: Dropped the P2M 1-1 setting]
[v3: Check for Xen in-case we are running under baremetal]
[v4: Wrap with CONFIG_PCI_MMCONFIG]
---
drivers/xen/pci.c | 47 +++++++++++++++++++++++++++++++++++++++++
include/xen/interface/physdev.h | 11 ++++++++++
2 files changed, 58 insertions(+)
diff --git a/drivers/xen/pci.c b/drivers/xen/pci.c
index 18fff88..d15f6e8 100644
--- a/drivers/xen/pci.c
+++ b/drivers/xen/pci.c
@@ -26,6 +26,7 @@
#include <asm/xen/hypervisor.h>
#include <asm/xen/hypercall.h>
#include "../pci/pci.h"
+#include <asm/pci_x86.h>
static bool __read_mostly pci_seg_supported = true;
@@ -192,3 +193,49 @@ static int __init register_xen_pci_notifier(void)
}
arch_initcall(register_xen_pci_notifier);
+
+#ifdef CONFIG_PCI_MMCONFIG
+static int __init xen_mcfg_late(void)
+{
+ struct pci_mmcfg_region *cfg;
+ int rc;
+
+ if (!xen_initial_domain())
+ return 0;
+
+ if ((pci_probe & PCI_PROBE_MMCONF) == 0)
+ return 0;
+
+ if (list_empty(&pci_mmcfg_list))
+ return 0;
+
+ /* Check whether they are in the right area. */
+ list_for_each_entry(cfg, &pci_mmcfg_list, list) {
+ struct physdev_pci_mmcfg_reserved r;
+
+ r.address = cfg->address;
+ r.segment = cfg->segment;
+ r.start_bus = cfg->start_bus;
+ r.end_bus = cfg->end_bus;
+ r.flags = XEN_PCI_MMCFG_RESERVED;
+
+ rc = HYPERVISOR_physdev_op(PHYSDEVOP_pci_mmcfg_reserved, &r);
+ switch (rc) {
+ case 0:
+ case -ENOSYS:
+ continue;
+
+ default:
+ pr_warn("Failed to report MMCONFIG reservation"
+ " state for %s to hypervisor"
+ " (%d)\n",
+ cfg->name, rc);
+ }
+ }
+ return 0;
+}
+/*
+ * Needs to be done after acpi_init which are subsys_initcall.
+ */
+subsys_initcall_sync(xen_mcfg_late);
+#endif
diff --git a/include/xen/interface/physdev.h b/include/xen/interface/physdev.h
index 7000bb1..42721d1 100644
--- a/include/xen/interface/physdev.h
+++ b/include/xen/interface/physdev.h
@@ -231,6 +231,17 @@ struct physdev_get_free_pirq {
#define XEN_PCI_DEV_VIRTFN 0x2
#define XEN_PCI_DEV_PXM 0x4
+#define XEN_PCI_MMCFG_RESERVED 0x1
+
+#define PHYSDEVOP_pci_mmcfg_reserved 24
+struct physdev_pci_mmcfg_reserved {
+ uint64_t address;
+ uint16_t segment;
+ uint8_t start_bus;
+ uint8_t end_bus;
+ uint32_t flags;
+};
+
#define PHYSDEVOP_pci_device_add 25
struct physdev_pci_device_add {
/* IN */
--
1.8.3.1
David Vrabel
2013-Nov-21 10:37 UTC
Re: [PATCH 1/4] xen/mcfg: Call PHYSDEVOP_pci_mmcfg_reserved for MCFG areas.
On 08/11/13 17:38, Konrad Rzeszutek Wilk wrote:> The PCI MMCONFIG area is usually reserved via the E820 so the Xen hypervisor > is aware of these regions. But they can also be enumerated in the ACPI > DSDT which means the hypervisor won''t know of them until the initial > domain informs it of via PHYSDEVOP_pci_mmcfg_reserved. > > This is what this patch does for all of the MCFG regions that the > initial domain is aware of (E820 enumerated and ACPI). > > Reported-by: Santosh Jodh <Santosh.Jodh@citrix.com> > CC: Jan Beulich <JBeulich@suse.com> > CC: Boris Ostrovsky <boris.ostrovsky@oracle.com> > CC: David Vrabel <david.vrabel@citrix.com> > CC: Mukesh Rathor <mukesh.rathor@oracle.com> > Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>Reviewed-by: David Vrabel <david.vrabel@citrix.com>> [v1: Redid it a bit] > [v2: Dropped the P2M 1-1 setting] > [v3: Check for Xen in-case we are running under baremetal] > [v4: Wrap with CONFIG_PCI_MMCONFIG]This sort of change list isn''t useful in the long term. I''d prefer it if this was after the --- divider.> --- a/drivers/xen/pci.c > +++ b/drivers/xen/pci.c > @@ -26,6 +26,7 @@ > + default: > + pr_warn("Failed to report MMCONFIG reservation" > + " state for %s to hypervisor" > + " (%d)\n",Message strings shouldn''t be split like this. David