iommu/vt-d: Set SSADE when attaching to a parent with dirty tracking

  Should set the SSADE (Second Stage Access/Dirty bit Enable) bit of the
pasid entry when attaching a device to a nested domain if its parent
has already enabled dirty tracking.

Fixes: 111bf85c68f6 ("iommu/vt-d: Add helper to setup pasid nested translation")
Signed-off-by: Yi Liu <yi.l.liu@intel.com>
---
base commit: 547ab8fc4cb04a1a6b34377dd8fad34cd2c8a8e3
---
 drivers/iommu/intel/pasid.c | 2 ++
 1 file changed, 2 insertions(+)
  

On 08/02/2024 09:14, Yi Liu wrote:
> Should set the SSADE (Second Stage Access/Dirty bit Enable) bit of the
> pasid entry when attaching a device to a nested domain if its parent
> has already enabled dirty tracking.
> 
> Fixes: 111bf85c68f6 ("iommu/vt-d: Add helper to setup pasid nested translation")
> Signed-off-by: Yi Liu <yi.l.liu@intel.com>

Reviewed-by: Joao Martins <joao.m.martins@oracle.com>

> ---
> base commit: 547ab8fc4cb04a1a6b34377dd8fad34cd2c8a8e3
> ---
>  drivers/iommu/intel/pasid.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/iommu/intel/pasid.c b/drivers/iommu/intel/pasid.c
> index 3239cefa4c33..9be24bb762cf 100644
> --- a/drivers/iommu/intel/pasid.c
> +++ b/drivers/iommu/intel/pasid.c
> @@ -658,6 +658,8 @@ int intel_pasid_setup_nested(struct intel_iommu *iommu, struct device *dev,
>  	pasid_set_domain_id(pte, did);
>  	pasid_set_address_width(pte, s2_domain->agaw);
>  	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
> +	if (s2_domain->dirty_tracking)
> +		pasid_set_ssade(pte);
>  	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_NESTED);
>  	pasid_set_present(pte);
>  	spin_unlock(&iommu->lock);
  

On 2024/2/8 17:14, Yi Liu wrote:
> Should set the SSADE (Second Stage Access/Dirty bit Enable) bit of the
> pasid entry when attaching a device to a nested domain if its parent
> has already enabled dirty tracking.
> 
> Fixes: 111bf85c68f6 ("iommu/vt-d: Add helper to setup pasid nested translation")
> Signed-off-by: Yi Liu <yi.l.liu@intel.com>
> ---
> base commit: 547ab8fc4cb04a1a6b34377dd8fad34cd2c8a8e3
> ---
>   drivers/iommu/intel/pasid.c | 2 ++
>   1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/iommu/intel/pasid.c b/drivers/iommu/intel/pasid.c
> index 3239cefa4c33..9be24bb762cf 100644
> --- a/drivers/iommu/intel/pasid.c
> +++ b/drivers/iommu/intel/pasid.c
> @@ -658,6 +658,8 @@ int intel_pasid_setup_nested(struct intel_iommu *iommu, struct device *dev,
>   	pasid_set_domain_id(pte, did);
>   	pasid_set_address_width(pte, s2_domain->agaw);
>   	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
> +	if (s2_domain->dirty_tracking)
> +		pasid_set_ssade(pte);
>   	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_NESTED);
>   	pasid_set_present(pte);
>   	spin_unlock(&iommu->lock);

The same fix should also be applied to the intel_pasid_setup_second()
path. Specifically, if a second-stage domain with dirty tracking enabled
attaches to a device, the corresponding entry in the pasid table should
also have the SSADE bit set.

Best regards,
baolu
  

On 09/02/2024 02:40, Baolu Lu wrote:
> On 2024/2/8 17:14, Yi Liu wrote:
>> Should set the SSADE (Second Stage Access/Dirty bit Enable) bit of the
>> pasid entry when attaching a device to a nested domain if its parent
>> has already enabled dirty tracking.
>>
>> Fixes: 111bf85c68f6 ("iommu/vt-d: Add helper to setup pasid nested translation")
>> Signed-off-by: Yi Liu <yi.l.liu@intel.com>
>> ---
>> base commit: 547ab8fc4cb04a1a6b34377dd8fad34cd2c8a8e3
>> ---
>>   drivers/iommu/intel/pasid.c | 2 ++
>>   1 file changed, 2 insertions(+)
>>
>> diff --git a/drivers/iommu/intel/pasid.c b/drivers/iommu/intel/pasid.c
>> index 3239cefa4c33..9be24bb762cf 100644
>> --- a/drivers/iommu/intel/pasid.c
>> +++ b/drivers/iommu/intel/pasid.c
>> @@ -658,6 +658,8 @@ int intel_pasid_setup_nested(struct intel_iommu *iommu,
>> struct device *dev,
>>       pasid_set_domain_id(pte, did);
>>       pasid_set_address_width(pte, s2_domain->agaw);
>>       pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
>> +    if (s2_domain->dirty_tracking)
>> +        pasid_set_ssade(pte);
>>       pasid_set_translation_type(pte, PASID_ENTRY_PGTT_NESTED);
>>       pasid_set_present(pte);
>>       spin_unlock(&iommu->lock);
> 
> The same fix should also be applied to the intel_pasid_setup_second()
> path. Specifically, if a second-stage domain with dirty tracking enabled
> attaches to a device, the corresponding entry in the pasid table should
> also have the SSADE bit set.
> 

I think that's already done, see below.


/*
 * Set up the scalable mode pasid entry for second only translation type.
 */
int intel_pasid_setup_second_level(struct intel_iommu *iommu,
				   struct dmar_domain *domain,
				   struct device *dev, u32 pasid)
{
(...)

	pasid_clear_entry(pte);
	pasid_set_domain_id(pte, did);
	pasid_set_slptr(pte, pgd_val);
	pasid_set_address_width(pte, agaw);
	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_SL_ONLY);
	pasid_set_fault_enable(pte);
	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
	if (domain->dirty_tracking)
		pasid_set_ssade(pte);
(...)
}
  

On 08/02/2024 10:31, Joao Martins wrote:
> On 08/02/2024 09:14, Yi Liu wrote:
>> Should set the SSADE (Second Stage Access/Dirty bit Enable) bit of the
>> pasid entry when attaching a device to a nested domain if its parent
>> has already enabled dirty tracking.
>>
>> Fixes: 111bf85c68f6 ("iommu/vt-d: Add helper to setup pasid nested translation")
>> Signed-off-by: Yi Liu <yi.l.liu@intel.com>
> 
> Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
> 
On a second thought, while the patch looks fine if this is what we wanna do,
just a quick clarification on the why (that's also applicable to the last patch
of your other series[0]). I am sure I am missing something :)

Shouldn't the nested domain be subdued to whatever features guest idea of
ecap/cap instead of host's view? Or is this because guest first-stage page table
on Intel is supposed to be always-enabled dirty tracking (per SDM) ? If it's the
latter, it probably should be sprinkled in the commit message(s).

[0] https://lore.kernel.org/linux-iommu/20240208082307.15759-9-yi.l.liu@intel.com/

>> ---
>> base commit: 547ab8fc4cb04a1a6b34377dd8fad34cd2c8a8e3
>> ---
>>  drivers/iommu/intel/pasid.c | 2 ++
>>  1 file changed, 2 insertions(+)
>>
>> diff --git a/drivers/iommu/intel/pasid.c b/drivers/iommu/intel/pasid.c
>> index 3239cefa4c33..9be24bb762cf 100644
>> --- a/drivers/iommu/intel/pasid.c
>> +++ b/drivers/iommu/intel/pasid.c
>> @@ -658,6 +658,8 @@ int intel_pasid_setup_nested(struct intel_iommu *iommu, struct device *dev,
>>  	pasid_set_domain_id(pte, did);
>>  	pasid_set_address_width(pte, s2_domain->agaw);
>>  	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
>> +	if (s2_domain->dirty_tracking)
>> +		pasid_set_ssade(pte);
>>  	pasid_set_translation_type(pte, PASID_ENTRY_PGTT_NESTED);
>>  	pasid_set_present(pte);
>>  	spin_unlock(&iommu->lock);
>
  

> From: Joao Martins <joao.m.martins@oracle.com>
> Sent: Friday, February 9, 2024 6:36 PM
> 
> On 08/02/2024 10:31, Joao Martins wrote:
> > On 08/02/2024 09:14, Yi Liu wrote:
> >> Should set the SSADE (Second Stage Access/Dirty bit Enable) bit of the
> >> pasid entry when attaching a device to a nested domain if its parent
> >> has already enabled dirty tracking.
> >>
> >> Fixes: 111bf85c68f6 ("iommu/vt-d: Add helper to setup pasid nested
> translation")
> >> Signed-off-by: Yi Liu <yi.l.liu@intel.com>
> >
> > Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
> >
> On a second thought, while the patch looks fine if this is what we wanna do,
> just a quick clarification on the why (that's also applicable to the last patch
> of your other series[0]). I am sure I am missing something :)
> 
> Shouldn't the nested domain be subdued to whatever features guest idea of
> ecap/cap instead of host's view? Or is this because guest first-stage page
> table
> on Intel is supposed to be always-enabled dirty tracking (per SDM) ? If it's the
> latter, it probably should be sprinkled in the commit message(s).
> 
> [0] https://lore.kernel.org/linux-iommu/20240208082307.15759-9-
> yi.l.liu@intel.com/
> 

first-stage dirty tracking is always enabled. but the real point here is
that the host has enabled dirty-tracking in second-stage so when 
a device is attached to a nested domain on top of that second-stage
we should set SSADE in the pasid entry for that device.

even if there is a guest ecap/cap for first-stage dirty tracking that
shouldn't affect the host setting for second-stage which is invisible
to the guest.