[RFC PATCH v1 00/18] Provide a new two step DMA API mapping API

[Chaitanya Kulkarni]

On 7/3/24 07:35, Christoph Hellwig wrote:
> On Wed, Jul 03, 2024 at 01:52:53PM +0300, Leon Romanovsky wrote:
>> On Wed, Jul 03, 2024 at 07:42:38AM +0200, Christoph Hellwig wrote:
>>> I just tried to boot this on my usual qemu test setup with emulated
>>> nvme devices, and it dead-loops with messages like this fairly late
>>> in the boot cycle:
>>>
>>> [   43.826627] iommu: unaligned: iova 0xfff7e000 pa 0x000000010be33650 size 0x1000 min_pagesz 0x1000
>>> [   43.826982] dma_mapping_error -12
>>>
>>> passing intel_iommu=off instead of intel_iommu=on (expectedly) makes
>>> it go away.
>> Can you please share your kernel command line and qemu?
>> On my and Chaitanya setups it works fine.
> qemu-system-x86_64 \
>          -nographic \
> 	-enable-kvm \
> 	-m 6g \
> 	-smp 4 \
> 	-cpu host \
> 	-M q35,kernel-irqchip=split \
> 	-kernel arch/x86/boot/bzImage \
> 	-append "root=/dev/vda console=ttyS0,115200n8 intel_iommu=on" \
>          -device intel-iommu,intremap=on \
> 	-device ioh3420,multifunction=on,bus=pcie.0,id=port9-0,addr=9.0,chassis=0 \	
>          -blockdev driver=file,cache.direct=on,node-name=root,filename=/home/hch/images/bookworm.img \
> 	-blockdev driver=host_device,cache.direct=on,node-name=test,filename=/dev/nvme0n1p4 \
> 	-device virtio-blk,drive=root \
> 	-device nvme,drive=test,serial=1234
>

I tried to reproduce this issue somehow it is not reproducible.

I'll try again on Leon's setup on my Saturday night, to fix that
case.

-ck