On 2023-03-28 02:22, Christoph Hellwig wrote:
I finally found some real time to look into this:
On Tue, Mar 21, 2023 at 02:52:00PM +0100, Linux regression tracking (Thorsten Leemhuis) wrote:
The man errors I am getting are
dmar_fault 8 callbacks suppressed
DMAR : DRHD: handling fault status req 2
DMAR : [DMA Write NO_PASID] Request device [07.00.1] fault addr
0xfffe0000 [fault reason 0x82] Present bit in contect entry is clear
This clearly indicates that my original idea about the AMD gart was
completely bonkers, as we're obviously on an Intel platform.
And this indicates that the device is trying to do a DMA write to
something that isn't IOMMU mapped. Getting this from an initialization
change (commit 78013eaadf6 (x86: remove the IOMMU table infrastructure")
feels very strange to me.
Can you maybe post the full dmesg? I wonder if there is interesting
initialization error in here.
FWIW "Marvell SATA" instantly makes me suspect the phantom function
quirk. What *should* happen is the IOMMU driver sees the PCI DMA aliases
correctly and sets up context entries for both 07.00.0 and 07.00.1, but
it looks like that may be what's gone awry.
Robin.
