Re: [PATCH 7/9] iommupt/x86: Support SW bits and permit PT_FEAT_DMA_INCOHERENT

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

On 7/30/25 07:13, Jason Gunthorpe wrote:
> On Tue, Jul 22, 2025 at 01:17:59PM +0800, Baolu Lu wrote:
>>> +static inline u64 x86_64_pt_sw_bit(unsigned int bitnr)
>>> +{
>>> +	/* Bits marked Ignored/AVL in the specification */
>>> +	switch (bitnr) {
>>> +	case 0:
>>> +		return BIT(9);
>>> +	case 1:
>>> +		return BIT(11);
>>> +	case 2 ... 12:
>>> +		return BIT_ULL((bitnr - 2) + 52);
>>
>> Just out of curiosity, why bit 63 can't be used as the sw bit?
> 
> It is not ignored, but used for XD/NX:
> 
> 	X86_64_FMT_XD = BIT_ULL(63),
> 
> Which should comes out for ATS replies.
> 
> The existing drivers didn't wire it up, so I left it alone as well,

The VT-d driver once had such support, but it was removed by commit
e995fcde6070 as the VT-d architecture doesn't support it anymore.

commit e995fcde6070f0981e083c1e2e17e401e6c17ad9
Author: Lu Baolu <baolu.lu@linux.intel.com>
Date:   Tue Jul 2 21:08:35 2024 +0800

     iommu/vt-d: Remove control over Execute-Requested requests

     The VT-d specification has removed architectural support of the 
requests
     with pasid with a value of 1 for Execute-Requested (ER). And the 
NXE bit
     in the pasid table entry and XD bit in the first-stage paging 
Entries are
     deprecated accordingly.

     Remove the programming of these bits to make it consistent with the 
spec.

     Suggested-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
     Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
     Reviewed-by: Kevin Tian <kevin.tian@intel.com>
     Link: 
https://lore.kernel.org/r/20240624032351.249858-1-baolu.lu@linux.intel.com
     Link: 
https://lore.kernel.org/r/20240702130839.108139-4-baolu.lu@linux.intel.com
     Signed-off-by: Will Deacon <will@kernel.org>


> but the XD should be linked to IOMMU_NOEXEC. There are some little
> things like this we can evaluate after. The idea here is to try hard
> to make no change to the drivers existing behavior.

Yeah, we can carve out that bit for now and evaluate it later, given
that we already have enough sw bits.

Thanks,
baolu