Re: [PATCH v6 00/20] dma-mapping: Use DMA_ATTR_CC_SHARED through direct, pool and swiotlb paths

[Alexey Kardashevskiy <aik@amd.com>]

On 10/6/26 00:47, Jason Gunthorpe wrote:
> On Tue, Jun 09, 2026 at 02:43:08PM +0100, Catalin Marinas wrote:
>> On Thu, Jun 04, 2026 at 02:09:39PM +0530, Aneesh Kumar K.V (Arm) wrote:
>>> This series propagates DMA_ATTR_CC_SHARED through the dma-direct,
>>> dma-pool, and swiotlb paths so that encrypted and decrypted DMA buffers
>>> are handled consistently.
>>>
>>> Today, the direct DMA path mostly relies on force_dma_unencrypted() for
>>> shared/decrypted buffer handling. This series consolidates the
>>> force_dma_unencrypted() checks in the top-level functions and ensures
>>> that the remaining DMA interfaces use DMA attributes to make the correct
>>> decisions.
>>
>> Please check Sashiko's reports, it has some good points:
>>
>> https://sashiko.dev/#/patchset/20260604083959.1265923-1-aneesh.kumar@kernel.org
>>
>> I think the main one is the swiotlb_tbl_map_single() changes which break
>> AMD SME host support. There cc_platform_has(CC_ATTR_MEM_ENCRYPT) is true
>> but force_dma_unencrypted() is false. Normally you'd not end up on this
>> path but you can have swiotlb=force.
> 
> IMHO that's an AMD issue, not with the design of this series..
> 
> The series is right, a device that is !force_dma_decrypted() must be
> considerd to be a trusted device and we must never place any DMA
> mappings for a trusted device into shared memory.


swiotlb=force forces swiotlb, not decryption.

> That AMD has done somethine insane:
> 
> bool force_dma_unencrypted(struct device *dev)
> {
> 	/*
> 	 * For SEV, all DMA must be to unencrypted addresses.
> 	 */
> 	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
> 		return true;
> 
> 	/*
> 	 * For SME, all DMA must be to unencrypted addresses if the
> 	 * device does not support DMA to addresses that include the
> 	 * encryption mask.
> 	 */
> 	if (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT)) {
> 		u64 dma_enc_mask = DMA_BIT_MASK(__ffs64(sme_me_mask));
> 		u64 dma_dev_mask = min_not_zero(dev->coherent_dma_mask,
> 						dev->bus_dma_limit);
> 
> 		if (dma_dev_mask <= dma_enc_mask)
> 			return true;
> 	}


So when I try "mem_encrypt=on iommu=pt swiotlb=force" with this patchset, it fails to boot. But it boots with a hack like this:

===
@@ -39,7 +41,7 @@ bool force_dma_unencrypted(struct device *dev)
                         return true;
         }
  
-       return false;
+       return swiotlb_force_bounce;
  }
===

Or we say "mem_encrypt=on iommu=pt swiotlb=force" combo is just weird and we won't be supporting which bit in this? Thanks,


> 
> Is an AMD issue. We already have an address mask limit system built
> into the DMA API, arch code should not be co-opting the CC mechanism
> to create a special pool for address limited devices.
> 
> The correct thing is to ensure the DMA API is checking any address
> limits on the actual true dma_addr_t, not on an intermediate like a
> phys_addr before it is adjusted with any C bit. Then it is a normal
> low address swiotlb bounce like any other.
> 
> I think we can ignore this Sashiko remark, in real systems the use of
> swiotlb for 64 bit devices is very rare. Though it would be good to
> remove this code from AMD...> 
> Jason

-- 
Alexey