Re: [BUG] iommu/dma: "Not yet supported" CC_SHARED rejection regresses dma_alloc_attrs() on SME hosts with IOMMU-attached GPUs (commit 8277a12d0d60)

[Aneesh Kumar K.V <aneesh.kumar@kernel.org>]

Marek Szyprowski <m.szyprowski@samsung.com> writes:

> On 02.09.2026 20:26, Deucher, Alexander wrote:
>>> -----Original Message-----
>>> From: Aneesh Kumar K.V <aneesh.kumar@kernel.org>
>>> Sent: Wednesday, September 2, 2026 6:16 AM
>>> To: Timo Witte <timo.witte@gmail.com>; iommu@lists.linux.dev
>>> Cc: m.szyprowski@samsung.com; robin.murphy@arm.com;
>>> joro@8bytes.org; Deucher, Alexander <Alexander.Deucher@amd.com>
>>> Subject: Re: [BUG] iommu/dma: "Not yet supported" CC_SHARED rejection
>>> regresses dma_alloc_attrs() on SME hosts with IOMMU-attached GPUs
>>> (commit 8277a12d0d60)
>>>
>>> Timo Witte <timo.witte@gmail.com> writes:
>>>
>>>> Commit 8277a12d0d60 ("dma-pool: track decrypted atomic pools and
>>>> select them via attrs") added to iommu_dma_alloc():
>>>>
>>>>     /* Not yet supported */
>>>>     if (is_alloc_cc_shared)
>>>>         return NULL;
>>>>
>>>> dma_alloc_attrs() auto-sets __DMA_ATTR_ALLOC_CC_SHARED whenever
>>>> force_dma_unencrypted() is true. On an SME host (mem_encrypt=on) an
>>>> AMD GPU with a 44-bit DMA mask hits that condition, so every GTT
>>>> (system-RAM) allocation fails with NULL/-ENOMEM, amdgpu init aborts
>>>> and the system boots without a display.
>>>> IOMMU in Translated/DMA-FQ mode (iommu group 29).
>>>>
>>>> Before the mentioned patch it worked: the IOMMU PTE carries the SME
>>>> C-bit, so the memory controller transparently decrypts for the device
>>>> while pages stay encrypted at rest. For IOMMU-backed devices
>>>> force_dma_unencrypted() is a false positive - the device DMA mask does
>>>> not limit addressing once the IOMMU translates, so CC_SHARED should
>>>> not be auto-set for them.
>>>>
>>>
>>> This means that, with host memory encryption, DMA does not need to be
>>> forced unencrypted when the device uses the IOMMU?
>> + Tom and Vasant.
>>
>> Yes, otherwise most devices won't be able to work with encrypted memory since most devices don't support a DMA mask large enough to cover the C bit.
>
> Is this the proper fix/workaround for v7.3-rc? If so, then Aneesh, please
> send it as a complete patch so I will take it to the dma-mapping-fixes.
>


I am still waiting for testing feedback. I sent a reminder on Friday.

-aneesh