Re: [BUG] VFIO on POWER9 / QEMU fails to allocate 32-bit DMA

Hi Timothy

Timothy Pearson [off-list ref] writes:

----- Original Message -----

quoted

From: "Ritesh Harjani" <ritesh.list@gmail.com>
To: "Timothy Pearson" <tpearson@raptorengineering.com>
Cc: "linuxppc-dev" <redacted>, "Shivaprasad G Bhat" <redacted>, "Gaurav Batra"
[off-list ref], "Madhavan Srinivasan" [off-list ref], "Venkat Rao Bagalkote" [off-list ref]
Sent: Saturday, August 8, 2026 9:13:09 AM
Subject: Re: [BUG] VFIO on POWER9 / QEMU fails to allocate 32-bit DMA

quoted

Timothy Pearson [off-list ref] writes:

quoted

----- Original Message -----
No worries, appreciated.  This is only one of several major problems with Linux
on PowerNV that we have run into after recent upgrades, and the primary focus
at the moment is on restoring broken functionality.  In many cases, that means
moving to different hardware, so gathering logs afterward can be difficult /
impossible.

Sorry to hear that you have been facing such issues recently. I will
bring this up internally. We will try to add more test matrix for
upstream PowerNV (note that we already have PowerNV tested regularly as
part of our upstream CI, but I suppose we could extend more KVM guest
testing there), so that we can reduce down on such reports.

That would be very helpful.  I have a number of other internal reports (and some external customer reports) that I will try to get together for submission here.  Given the testing on PowerNV seems quite sparse at the moment, if you do need additional PowerNV machines for testing we have bare metal cloud leases available.

Thanks for bringing up your concern. Based on your feedback, we brought
up this topic internally. So currently w.r.t upstream maintainance work,
we have build, boot, selftests and few other test runs, on our PowerNV
systems, but I agree our other major testing was happening on Pseries
platform. So, we decided to add more PowerNV machines to our internal CI
and expand / increase our upstream CI tests and runs on these baremetal
machines too. This is going to happen very soon, so hopefully, this
should reduce such reports in the future.


BTW - notice that the current issue - I reproduced on Pseries
development environment itself. So this one seems to be some special
case where you either have a 32-bit and a 64-bit HW sitting on the same
PE (partitionable endpoint) behind the same iommu group (like sometimes
behind a PCIe switch) or maybe the HW does both 32-bit and 64-bit probe.
But anyways the patch I pointed should fix that problem and we will
ensure it is backported to all stable kernels, and likely the
distributions should pick that up as well (we will take care of that too).

One of the customer reports I already forwarded up to Github, but further logs aren't available as the customer has already moved to different hardware:

https://github.com/tbsdtv/linux_media/issues/431

The final straw that caused migration on that side was yet another VFIO bug on top of the invalid DMA that prevented EEH recovery and was forcing host resets to thaw the frozen PE.  I'll try to submit that report here as well.

Sure thanks. I see you already have shared few reports. We will take a
look at those. If there is anything else please feel free to share it
with us.

quoted

Although not with VFIO - but I verified this issue by adding e1000
32-bit DMA device on the same PHB where other 64bit DMA devices were
attached (in my Qemu development environment).

And with that I was able to re-create the issue with kexec i.e. on kexec
I see the following error similar to what you reported in your
environment:

[    5.537726] pci 0000:00:01.0: ibm,query-pe-dma-windows(2026) 800 8000000
20000000 returned 0, lb=2000000000 ps=107 wn=0
[    5.540233] pci 0000:00:01.0: Adding to iommu group 0
[    5.556968] pci 0000:00:03.0: Adding to iommu group 0
[    5.567164] PCI: Probing PCI hardware done
[    9.395499] virtio-pci 0000:00:03.0: lsa_required: 0, lsa_enabled: 0, direct
mapping: 0
[    9.396111] virtio-pci 0000:00:03.0: lsa_required: 0, lsa_enabled: 0, direct
mapping: 0
[   13.228575] e1000: Intel(R) PRO/1000 Network Driver
[   13.229566] e1000: Copyright (c) 1999-2006 Intel Corporation.
[   13.253804] e1000 0000:00:01.0: Warning: IOMMU offset too big for device mask
[   13.255200] e1000 0000:00:01.0: mask: 0xffffffff, table offset:
0x800000000000000
[   13.256611] e1000: No usable DMA config, aborting
[   13.267336] e1000 0000:00:01.0: probe with driver e1000 failed with error -5

This is failing with same error as you had reported:
    "Warning: IOMMU offset too big for device mask"

Whereas after I applied the patch mentioned in [1], I am able to see the
device working fine after kexec. Here are the logs with the fix applied:

[    6.197623] pci 0000:00:01.0: ibm,query-pe-dma-windows(2026) 800 8000000
20000000 returned 0, lb=2000000000 ps=107 wn=0
[    6.201686] pci 0000:00:01.0: Adding to iommu group 0
[    6.221890] pci 0000:00:03.0: Adding to iommu group 0
[    8.991990] virtio-pci 0000:00:03.0: lsa_required: 0, lsa_enabled: 0, direct
mapping: 0
[    8.992710] virtio-pci 0000:00:03.0: lsa_required: 0, lsa_enabled: 0, direct
mapping: 0
[   11.315607] e1000: Intel(R) PRO/1000 Network Driver
[   11.316191] e1000: Copyright (c) 1999-2006 Intel Corporation.
[   11.671185] e1000 0000:00:01.0 eth1: (PCI:33MHz:32-bit) 52:54:00:12:34:56
[   11.672604] e1000 0000:00:01.0 eth1: Intel(R) PRO/1000 Network Connection

quoted

Thank you for the link.  When / if we want to test a migration of that hardware
back to VFIO on PowerNV I will ensure that patch is applied before testing.

Sure, I think this should fix the issue you reported. We have tagged
that fix with stable - so we will ensure it is backported to stable
kernel releases as well.

That sounds great, thanks.  Does IBM have any plans to support Debian LTS (Freexian) so these patches can flow into the broken distribution kernels?

yes, we do support that. Once the patch is merged and available in
stable kernels, we will ensure this is backported to this distribution
as well.

Thanks!

I see you have already shared few more reports with us. We will be
looking into those and will bring those to closure too.

Thanks again for sharing the details with us.

-ritesh