Re: [PATCH 1/8] iommu: Lift and generalize the STE/CD update code from SMMUv3

[Baolu Lu <baolu.lu@linux.intel.com>]

On 3/23/26 20:59, Jason Gunthorpe wrote:
> On Mon, Mar 16, 2026 at 02:24:57PM +0800, Baolu Lu wrote:
> 
>>>> +	writer->ops->get_used(entry, cur_used);
>>>> +	writer->ops->get_used(target, target_used);
>>>
>>> SMMU has get_update_safe now. Can we take it together?
>>
>> I will look into the SMMUv3 get_update_safe implementation. Or integrate
>> that specially when we transition the ARM SMMUv3 driver to use this
>> generic entry_sync library.
> 
> The intention was to copy the existing ARM code as is, the draft I
> sent was before these changes from Nicolin, so it should get updated..

Okay.

> 
>>>> +EXPORT_SYMBOL(NS(entry_sync_write));
>>>
>>> There is also a KUNIT test coverage in arm-smmu-v3 for all of these
>>> functions. Maybe we can make that generic as well?
>>
>> Same here.
> 
> That will be a bit hard since it depends on driver functions.
> 
>>> But maybe we could just call them:
>>>       entry_sync_writer_le64
>>>       entry_sync_writer_u128
>>> ?
>> I'm fine with the new naming. It is more explicit. I will update the
>> names unless there are further objections.
> 
> I was wondering if we should just be using void * here as the type
> safety seems a bit harmful if the goal is to make the 128 bit option
> fall back to 64 bits if not supported.
> 
> The maximum supported HW atomic quanta can be passed in through the
> struct.

I will explore refactoring the library to use void * and a dynamic
quanta size for v2.

Thanks,
baolu