[Qemu-devel] [PULL 14/17] hw/pci-bridge: disable SHPC in PXB

["Michael S. Tsirkin" <mst@redhat.com>]

From: Laszlo Ersek <lersek@redhat.com>

OVMF downloads the ACPI linker/loader script from QEMU when the edk2 PCI
Bus driver globally signals the firmware that PCI enumeration and resource
allocation have completed. At this point QEMU regenerates the ACPI payload
in an fw_cfg read callback, and this is when the PXB's _CRS gets
populated.

Unfortunately, when this happens, the PCI_COMMAND_MEMORY bit is clear in
the root bus's command register, *unlike* under SeaBIOS. The consequences
unfold as follows:

- When build_crs() fetches dev->io_regions[i].addr, it is all-bits-one,
  because pci_update_mappings() --> pci_bar_address() calculated it as
  PCI_BAR_UNMAPPED, due to the PCI_COMMAND_MEMORY bit being clear.

- Consequently, the SHPC MMIO BAR (bar 0) of the bridge is not added to
  the _CRS, *despite* having been programmed in PCI config space.

- Similarly, the SHPC MMIO BAR of the PXB is not removed from the main
  root bus's DWordMemory descriptor.

- Guest OSes (Linux and Windows alike) notice the pre-programmed SHPC BAR
  within the PXB's config space, and notice that it conflicts with the
  main root bus's memory resource descriptors. Linux reports

  pci 0000:04:00.0: BAR 0: can't assign mem (size 0x100)
  pci 0000:04:00.0: BAR 0: trying firmware assignment [mem
                           0x88200000-0x882000ff 64bit]
  pci 0000:04:00.0: BAR 0: [mem 0x88200000-0x882000ff 64bit] conflicts
                           with PCI Bus 0000:00 [mem
                           0x88200000-0xfebfffff]

  While Windows Server 2012 R2 reports

    https://technet.microsoft.com/en-us/library/cc732199%28v=ws.10%29.aspx

    This device cannot find enough free resources that it can use. If you
    want to use this device, you will need to disable one of the other
    devices on this system. (Code 12)

This issue was apparently encountered earlier, see the "hack" in:

  https://lists.nongnu.org/archive/html/qemu-devel/2015-01/msg02983.html

and the current hole-punching logic in build_crs() and build_ssdt() is
probably supposed to remedy exactly that problem -- however, for OVMF they
don't work, because at the end of the PCI enumeration and resource
allocation, which cues the ACPI linker/loader client, the command register
is clear.

The "shpc" property of "pci-bridge", introduced in the previous patches,
allows us to disable the standard hotplug controller cleanly, eliminating
the SHPC bar and the conflict.

Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Marcel Apfelbaum <marcel@redhat.com>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
---
 hw/pci-bridge/pci_expander_bridge.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/hw/pci-bridge/pci_expander_bridge.c b/hw/pci-bridge/pci_expander_bridge.c
index 3d840ef..70708ef 100644
--- a/hw/pci-bridge/pci_expander_bridge.c
+++ b/hw/pci-bridge/pci_expander_bridge.c
@@ -177,6 +177,7 @@ static int pxb_dev_initfn(PCIDevice *dev)
     bds = qdev_create(BUS(bus), "pci-bridge");
     bds->id = dev_name;
     qdev_prop_set_uint8(bds, PCI_BRIDGE_DEV_PROP_CHASSIS_NR, pxb->bus_nr);
+    qdev_prop_set_bit(bds, PCI_BRIDGE_DEV_PROP_SHPC, false);
 
     PCI_HOST_BRIDGE(ds)->bus = bus;
 
-- 
MST