NVIDIA: SAUCE: iommu/arm-smmu-v3: Use identity domain for ASPEED BMC devices by kobak2026 · Pull Request #371 · NVIDIA/NV-Kernels

kobak2026 · 2026-04-16T16:39:16Z

[Description]

ASPEED BMC devices (VGA, USB) behind an AST1150 PCIe-to-PCI bridge receive
DMA from BMC firmware using host physical addresses that bypass the kernel's
DMA API. When these devices are assigned a DMA translated domain, the SMMU
generates F_TRANSLATION faults because the BMC's physical addresses have no
corresponding IOVA mappings in the SMMU page tables.

Without the AST1150 NO_ALIAS quirk (commit 550a190), USB probe failed
in arm_smmu_insert_master() due to StreamID collision with VGA, leaving USB
in SMMU bypass mode and masking the BMC DMA issue. With the quirk applied,
both devices probe successfully into a translated domain, exposing the
underlying BMC DMA problem.

Fix this by returning IOMMU_DOMAIN_IDENTITY in arm_smmu_def_domain_type()
for PCI devices whose parent bridge has both the PCI_BRIDGE_NO_ALIASES flag
and an ASPEED vendor ID. Identity domain passes BMC DMA through untranslated.

Bug: 5918716

[How to verify and results]

Platform: Vera
Kernel: 6.17.13-kobak-no64dma+ based on 24.04_linux-nvidia-6.17-next
Boot cmdline: iommu.passthrough=0

Boot without patch, check dmesg:

$ sudo dmesg | grep F_TRANSLATION

Expected: 8+ F_TRANSLATION faults for client 0008:02:02.0 (ASPEED USB)
Apply patch, rebuild, reboot, check dmesg:

$ sudo dmesg | grep F_TRANSLATION

Expected: Zero F_TRANSLATION for ASPEED devices
Verify USB and VGA functional:

$ lsusb # ASPEED virtual hub visible
$ sudo dmesg | grep cdc_ether # CDC Ethernet registered
$ sudo dmesg | grep ast.*drm # AST DRM loaded

Results:

Test	ASPEED F_TRANSLATION	USB	VGA
Without fix (passthrough=0)	8+ faults per boot	Works	Works
iommu.passthrough=1	Zero	Works	Works
With identity domain patch	Zero	Works	Works

Verified by:

Koba Ko: 2026-04-15 on Vera
Carol Soto: 2026-04-15 on Vera eboard — "the patch is ok"

LP: https://bugs.launchpad.net/ubuntu/+source/linux-nvidia/+bug/2150470

github-actions · 2026-04-16T16:53:04Z

✅ Patchscan: No Missing Fixes

All cherry-picked commits have been checked — no missing upstream fixes found.

clsotog

Acked-by: Carol L Soto <csoto@nvidia.com>

nirmoy · 2026-04-16T17:38:04Z

+		    (pdev->bus->self->dev_flags &
+		     PCI_DEV_FLAGS_PCI_BRIDGE_NO_ALIASES) &&
+		    pdev->bus->self->vendor == PCI_VENDOR_ID_ASPEED)
+			return IOMMU_DOMAIN_IDENTITY;


This will hide future ASPEED firmware bug we should do affected device ID check like AST1150 device ID and report this to ASPEED.

FWDed an email which has contact of aspeed folks. Is this regression hapened after the patch which added PCI_DEV_FLAGS_PCI_BRIDGE_NO_ALIASES ? PCI_DEV_FLAGS_PCI_BRIDGE_NO_ALIASES actually fixed a real issue it was working on Grace and Vera. So I wonder if there was firmware/HWE kernel update that introduced this regression.

okay, ack.
yes, after applied, the issue is observed in lego and vera
do you know the aspeed contacts?

I have forwarded you an email discussion that I had for PCI_DEV_FLAGS_PCI_BRIDGE_NO_ALIASES

Did you not observe the F_TRANSLATION ?

This will hide future ASPEED firmware bug we should do affected device ID check like AST1150 device ID and report this to ASPEED.

This will hide future ASPEED firmware bug we should do affected device ID check like AST1150 device ID and report this to ASPEED.

Added device check.

There was one F_TRANSLATION at the boot time but lsusb showed all the usb devices properly. That time I didn't think F_TRANSLATION was related to the usb controller but now I think it was.

I thought the regression happened with that PCI_DEV_FLAGS_PCI_BRIDGE_NO_ALIASES but at the lego system I went back to 6.14 kernel and with iommu.passthrough=0 I will see the translation errors. If I use iommu.passthrough=1 I do not see it.
At vera eboard when I tried iommu.passthrough=1 last month, I was seeing another error but last night did not see it.

I think PCI_DEV_FLAGS_PCI_BRIDGE_NO_ALIASES just exposed this firmware issue but I am not sure why I didn't see this

nirmoy

Acked-by: Nirmoy Das <nirmoyd@nvidia.com>

nvmochs · 2026-04-16T19:08:06Z

No issues with the patch.

Acked-by: Matthew R. Ochs <mochs@nvidia.com>

@KobaKoNvidia - What are your plans for upstreaming this? Do we need to discuss with ASPEED first if they can (or plan to) address in firmware?

clsotog · 2026-04-16T19:46:54Z

Retested. Acked-by: Carol L Soto <csoto@nvidia.com>

kobak2026 · 2026-04-17T02:31:25Z

No issues with the patch.

Acked-by: Matthew R. Ochs <mochs@nvidia.com>

@KobaKoNvidia - What are your plans for upstreaming this? Do we need to discuss with ASPEED first if they can (or plan to) address in firmware?

I sent mail to ASPEED guy so let's have a discussion first

…devices ASPEED BMC devices behind an AST1150 PCIe-to-PCI bridge receive DMA from BMC firmware using host physical addresses that bypass the kernel's DMA API entirely. When these devices are assigned a DMA translated domain, the SMMU generates F_TRANSLATION faults because the BMC's physical addresses have no corresponding IOVA mappings in the SMMU page tables. Fix this by returning IOMMU_DOMAIN_IDENTITY for PCI devices whose parent bridge has both the PCI_BRIDGE_NO_ALIASES flag and an ASPEED vendor ID, so the SMMU passes BMC DMA transactions through untranslated. Signed-off-by: Koba Ko <kobak@nvidia.com>

github-actions · 2026-04-27T16:08:08Z

PR Validation Report

Patchscan ✅ No Missing Fixes

All cherry-picked commits checked — no missing upstream fixes found.

PR Lint ⚠️ Warnings

Details

Checking 1 commits...

Cherry-pick digest:
┌──────────────┬───────────────────────────────────────────────┬────────────┬─────────┬───────────────────────────┐
│ Local        │ Referenced upstream / Patch subject           │ Patch-ID   │ Subject │ SoB chain                 │
├──────────────┼───────────────────────────────────────────────┼────────────┼─────────┼───────────────────────────┤
│ 55f7351fc1d1 │ [SAUCE] iommu/arm-smmu-v3: use identity domai │ N/A        │ N/A     │ kobak                     │
└──────────────┴───────────────────────────────────────────────┴────────────┴─────────┴───────────────────────────┘

Lint: all checks passed.

PR metadata:
W: PR title missing [<branch>] prefix: "NVIDIA: SAUCE: iommu/arm-smmu-v3: Use identity domain for ASPEED BMC devices"

nvmochs · 2026-04-27T16:30:17Z

Merged, closing PR.

738fff0e2060 (nnoble/nvidia-6.17-next) NVIDIA: SAUCE: iommu/arm-smmu-v3: Use identity domain for ASPEED BMC devices

kobak2026 force-pushed the bug-5918716/smmu-identity-aspeed branch from 405bb46 to 53b7614 Compare April 16, 2026 16:43

kobak2026 requested review from clsotog, nirmoy and nvmochs April 16, 2026 16:43

clsotog approved these changes Apr 16, 2026

View reviewed changes

nirmoy reviewed Apr 16, 2026

View reviewed changes

kobak2026 force-pushed the bug-5918716/smmu-identity-aspeed branch from 53b7614 to a21698b Compare April 16, 2026 17:57

nirmoy approved these changes Apr 16, 2026

View reviewed changes

nvidia-bfigg force-pushed the 24.04_linux-nvidia-6.17-next branch from 8b07926 to 80bac29 Compare April 21, 2026 12:02

This was referenced Apr 27, 2026

[for 26.04_linux-nvidia]: NVIDIA: SAUCE: iommu/arm-smmu-v3: Use identity domain for ASPEED BMC devices #391

Closed

[for 26.04_linux-nvidia-bos]: NVIDIA: SAUCE: iommu/arm-smmu-v3: Use identity domain for ASPEED BMC devices #392

Closed

kobak2026 force-pushed the bug-5918716/smmu-identity-aspeed branch from a21698b to 55f7351 Compare April 27, 2026 15:59

nvmochs closed this Apr 27, 2026

Conversation

kobak2026 commented Apr 16, 2026 • edited by nvmochs Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Apr 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ Patchscan: No Missing Fixes

Uh oh!

clsotog left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kobak2026 Apr 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

nirmoy left a comment

Choose a reason for hiding this comment

Uh oh!

nvmochs commented Apr 16, 2026

Uh oh!

clsotog commented Apr 16, 2026

Uh oh!

kobak2026 commented Apr 17, 2026

Uh oh!

github-actions Bot commented Apr 27, 2026

PR Validation Report

Patchscan ✅ No Missing Fixes

PR Lint ⚠️ Warnings

Uh oh!

nvmochs commented Apr 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

kobak2026 commented Apr 16, 2026 •

edited by nvmochs

Loading

github-actions Bot commented Apr 16, 2026 •

edited

Loading

kobak2026 Apr 16, 2026 •

edited

Loading