Default arm virt machine broken

[Jonathan Marler]

This occurs on qemu_v3.0.0 but not on qemu_v2.12.2 (built from qemu_v3.0.0 tag on github)

Symptom: You'll see something like this in the kernel output:

[ 1.285210] OF: PCI: host bridge /pcie@10000000 ranges:
[ 1.286246] OF: PCI: IO 0x3eff0000..0x3effffff -> 0x00000000
[ 1.287061] OF: PCI: MEM 0x10000000..0x3efeffff -> 0x10000000
[ 1.287820] OF: PCI: MEM 0x8000000000..0xffffffffff -> 0x8000000000
[ 1.289312] pci-host-generic 4010000000.pcie: can't claim ECAM area [mem 0x10000000-0x1fffffff]: address conflict with /pcie@10000000 [mem 0x10000000-0x3efeffff]
[ 1.290984] pci-host-generic: probe of 4010000000.pcie failed with error -16

Qemu Command Line: qemu-system-arm -machine virt -m 1024M -kernel zImage -serial stdio

I can post my zImage if anyone has problems reproducing with their own zImage.

Note that this problem breaks pci making the machine unusable.

[Jonathan Marler (marler8997)]

Oh, I forgot, I should have also posted the relevant DTS contents:

 pcie@10000000 {
  interrupt-map-mask = <0x1800 0x0 0x0 0x7>;
  interrupt-map = <0x0 0x0 0x0 0x1 0x8001 0x0 0x0 0x0 0x3 0x4 0x0 0x0 0x0 0x2 0x8001 0x0 0x0 0x0 0x4 0x4 0x0 0x0 0x0 0x3 0x8001 0x0 0x0 0x0 0x5 0x4 0x0 0x0 0x0 0x4 0x8001 0x0 0x0 0x0 0x6 0x4 0x800 0x0 0x0 0x1 0x8001 0x0 0x0 0x0 0x4 0x4 0x800 0x0 0x0 0x2 0x8001 0x0 0x0 0x0 0x5 0x4 0x800 0x0 0x0 0x3 0x8001 0x0 0x0 0x0 0x6 0x4 0x800 0x0 0x0 0x4 0x8001 0x0 0x0 0x0 0x3 0x4 0x1000 0x0 0x0 0x1 0x8001 0x0 0x0 0x0 0x5 0x4 0x1000 0x0 0x0 0x2 0x8001 0x0 0x0 0x0 0x6 0x4 0x1000 0x0 0x0 0x3 0x8001 0x0 0x0 0x0 0x3 0x4 0x1000 0x0 0x0 0x4 0x8001 0x0 0x0 0x0 0x4 0x4 0x1800 0x0 0x0 0x1 0x8001 0x0 0x0 0x0 0x6 0x4 0x1800 0x0 0x0 0x2 0x8001 0x0 0x0 0x0 0x3 0x4 0x1800 0x0 0x0 0x3 0x8001 0x0 0x0 0x0 0x4 0x4 0x1800 0x0 0x0 0x4 0x8001 0x0 0x0 0x0 0x5 0x4>;
  #interrupt-cells = <0x1>;
  ranges = <0x1000000 0x0 0x0 0x0 0x3eff0000 0x0 0x10000 0x2000000 0x0 0x10000000 0x0 0x10000000 0x0 0x2eff0000 0x3000000 0x80 0x0 0x80 0x0 0x80 0x0>;
  reg = <0x40 0x10000000 0x0 0x10000000>;
  msi-parent = <0x8002>;
  dma-coherent;
  bus-range = <0x0 0xff>;
  linux,pci-domain = <0x0>;
  #size-cells = <0x2>;
  #address-cells = <0x3>;
  device_type = "pci";
  compatible = "pci-host-ecam-generic";
 };

[Jonathan Marler (marler8997)]

I tried to triage this a bit today.

I'm running a 32-bit linux kernel and I think that's the problem. The ECAM address base is at 0x4010000000, but it gets truncated to 0x10000000 because it's only a 32-bit kernel, but since it's truncated, it conflicts with VIRT_PCIE_MMIO (see hw/arm/virt.c) whose range is from 0x10000000 to 0x3efeffff which matches what we see in the error message.

[Eric Auger (augere)]

Hi Jonathan,

I sent an email yesterday on the qemu ML.

"
Please can you try using
qemu-system-arm -machine virt,highmem=off -m 1024M -kernel zImage
-serial stdio

Does your guest support LPAE? This may be the cause.

Thanks

Eric
"

[Jonathan Marler (marler8997)]

LPAE is actually disabled in my kernel config. Knowing the cause now, I can see that qemu would not be able to detect this problem. This error should have been detected in the linux kernel with an indication that the ECAM window was using a 40-bit address but LPAE was not enabled.