Unlike Linux on x86, which typically honors the PCI configuration performed
by the firmware, Linux on ARM assumes that the PCI subsystem needs to be
configured from scratch. This is not entirely unreasonable given the
historical background of embedded systems using very basic bootloaders,
but is no longer tenable with Linux on arm64 moving to UEFI and ACPI in the
server space. For this reason, PCI support in the arm64 kernel running under
ACPI is likely to move to the x86 model of honoring the PCI configuration
done by the firmware.

So let's align with that in our DT based configuration as well, and set the
/chosen/linux,pci-probe-only property to 1 in the Device Tree before we
hand it to the OS.

In case we are exposing an emulated VGA PCI device to the guest, which may
subsequently get exposed via the Graphics Output protocol and driven as an
efifb by the OS, this ensures the PCI resource allocations for the framebuffer
are not overridden, since that would cause the framebuffer to stop working.

Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Ard Biesheuvel <ard.biesheu...@linaro.org>
---
 ArmVirtPkg/VirtFdtDxe/VirtFdtDxe.c | 23 +++++++++++++++++++-
 1 file changed, 22 insertions(+), 1 deletion(-)

diff --git a/ArmVirtPkg/VirtFdtDxe/VirtFdtDxe.c 
b/ArmVirtPkg/VirtFdtDxe/VirtFdtDxe.c
index 74f80d1d2b78..36484a0bbb7e 100644
--- a/ArmVirtPkg/VirtFdtDxe/VirtFdtDxe.c
+++ b/ArmVirtPkg/VirtFdtDxe/VirtFdtDxe.c
@@ -286,6 +286,7 @@ InitializeVirtFdtDxe (
   VOID                           *DeviceTreeBase;
   INT32                          Node, Prev;
   INT32                          RtcNode;
+  INT32                          ChosenNode;
   EFI_STATUS                     Status;
   CONST CHAR8                    *Type;
   INT32                          Len;
@@ -356,8 +357,28 @@ InitializeVirtFdtDxe (
       ASSERT (Len == 2 * sizeof (UINT64));
       Status = ProcessPciHost (DeviceTreeBase, Node, RegProp);
       ASSERT_EFI_ERROR (Status);
-      break;
 
+      //
+      // Set the /chosen/linux,pci-probe-only property to 1, so that the PCI
+      // setup we will perform in the firmware is honored by the Linux OS,
+      // rather than torn down and done from scratch. This is generally a more
+      // sensible approach, and aligns with what ACPI based OSes do in general.
+      //
+      // In case we are exposing an emulated VGA PCI device to the guest, which
+      // may subsequently get exposed via the Graphics Output protocol and
+      // driven as an efifb by Linux, we need this setting to prevent the
+      // framebuffer from becoming unresponsive.
+      //
+      ChosenNode = fdt_path_offset (DeviceTreeBase, "/chosen");
+      if (ChosenNode < 0) {
+        ChosenNode = fdt_add_subnode (DeviceTreeBase, 0, "/chosen");
+      }
+      if (ChosenNode < 0) {
+        DEBUG ((EFI_D_WARN, "Failed to set /chosen/linux,pci-probe-only 
property"));
+        break;
+      }
+      fdt_setprop_u32 (DeviceTreeBase, ChosenNode, "linux,pci-probe-only", 1);
+      break;
     case PropertyTypeFwCfg:
       ASSERT (Len == 2 * sizeof (UINT64));
 
-- 
1.9.1

_______________________________________________
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel

Reply via email to