[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-ppc] PCI resource alignment differs for hotplugged devices

From: Shawn Anastasio
Subject: Re: [Qemu-ppc] PCI resource alignment differs for hotplugged devices
Date: Wed, 15 May 2019 19:55:27 -0500

After some digging, I have discovered that custom PCI resource
alignment is disabled on pseries in the kernel. I have submitted
a patch series to LKML that is still pending feedback that
addresses the issue: https://patchwork.ozlabs.org/cover/1096776/.

If anybody would like to review it or provide feedback, it would be
greatly appreciated.

- Shawn

On 5/6/19 6:21 PM, Shawn Anastasio wrote:
Hello all,

When hotplugging QEMU ivshmem-doorbell devices on a ppc64 pseries guest,
it seems that the assigned resource for BAR 0 is on a
4K page boundary, but not 64k:

[   25.606930] pci 0001:00:00.0: BAR 2: assigned [mem 0x220000000000-0x2200000fffff 64bit pref] [   25.615257] pci 0001:00:00.0: BAR 1: assigned [mem 0x200100000000-0x200100000fff] [   25.617284] pci 0001:00:00.0: BAR 0: assigned [mem 0x200100001000-0x2001000010ff]

This means that the device is not usable with the VFIO framework
on kernels with a 64k page size. After rebooting, BAR 0's address
is 64k-aligned and can be accessed like usual.

Is there any way to indicate to the system that BAR 0 should
be 64k aligned?

Thanks in advance,

reply via email to

[Prev in Thread] Current Thread [Next in Thread]