Re: [PATCH v2 2/3] iommu/vt-d: Check FL and SL capability sanity in scalable mode

[Lu Baolu]

On 9/27/21 9:33 AM, Tian, Kevin wrote:
> > From: Tian, Kevin
> > Sent: Monday, September 27, 2021 9:30 AM
> >
> > > From: Tian, Kevin
> > > Sent: Monday, September 27, 2021 9:12 AM
> > >
> > > > From: Lu Baolu <baolu.lu@xxxxxxxxxxxxxxx>
> > > > Sent: Sunday, September 26, 2021 7:46 PM
> > > >
> > > > An iommu domain could be allocated and mapped before it's attached to
> > > > any
> > > > device. This requires that in scalable mode, when the domain is allocated,
> > > > the format (FL or SL) of the page table must be determined. In order to
> > > > achieve this, the platform should support consistent SL or FL capabilities
> > > > on all IOMMU's. This adds a check for this and aborts IOMMU probing if it
> > > > doesn't meet this requirement.
> > >
> > > Is this a must? Looks the requirement comes from how the current code
> > > is implemented. It sets DOMAIN_FLAG_USE_FIRST_LEVEL flag in
> > > alloc_domain. But actually the pgtable is not allocated until the 1st device
> > > is attached. If this understanding is correct, you can also postpone the flag
> > > setting until pgtable is actually allocated.
> >
> > Baolu explained to me that RMRR regions are mapped before device
> > attach. So this check is necessary
> >
> > > of course how to handle inconsistent IOMMU capabilities is another
> > > orthogonal problem. Addressing it should not be only applied to SL/FL
> > > difference. especially this patch doesn't check consistency. it just
> > > checks that an IOMMU must support either SL or FL which doesn't
> > > match the commit msg here.
> >
> > and the overall inconsistency check mechanism is already in place.
> > and the logic here just extends it to cover SL/FL. Given that,
> >
> > Reviewed-by: Kevin Tian <kevin.tian@xxxxxxxxx>
> >
> >
> > > > Signed-off-by: Lu Baolu <baolu.lu@xxxxxxxxxxxxxxx>
> > > > ---
> > > >   drivers/iommu/intel/cap_audit.h |  1 +
> > > >   drivers/iommu/intel/cap_audit.c | 13 +++++++++++++
> > > >   2 files changed, 14 insertions(+)
> > > >
> > > > diff --git a/drivers/iommu/intel/cap_audit.h
> > > > b/drivers/iommu/intel/cap_audit.h
> > > > index 74cfccae0e81..d07b75938961 100644
> > > > --- a/drivers/iommu/intel/cap_audit.h
> > > > +++ b/drivers/iommu/intel/cap_audit.h
> > > > @@ -111,6 +111,7 @@ bool intel_cap_smts_sanity(void);
> > > >   bool intel_cap_pasid_sanity(void);
> > > >   bool intel_cap_nest_sanity(void);
> > > >   bool intel_cap_flts_sanity(void);
> > > > +bool intel_cap_slts_sanity(void);
> > > >
> > > >   static inline bool scalable_mode_support(void)
> > > >   {
> > > > diff --git a/drivers/iommu/intel/cap_audit.c
> > > > b/drivers/iommu/intel/cap_audit.c
> > > > index b12e421a2f1a..040e4ae0e42b 100644
> > > > --- a/drivers/iommu/intel/cap_audit.c
> > > > +++ b/drivers/iommu/intel/cap_audit.c
> > > > @@ -163,6 +163,14 @@ static int cap_audit_static(struct intel_iommu
> > > > *iommu, enum cap_audit_type type)
> > > >   			check_irq_capabilities(iommu, i);
> > > >   	}
> > > >
> > > > +	/*
> > > > +	 * If the system is sane to support scalable mode, either SL or FL
> > > > +	 * should be sane.
> > > > +	 */
> > > > +	if (intel_cap_smts_sanity() &&
> > > > +	    !intel_cap_flts_sanity() && !intel_cap_slts_sanity())
> > > > +		return -EFAULT;
>
> btw this should not be -EFAULT.

Agreed. I will change it to -ENOTSUPP.

Best regards,
baolu

> > > > +
> > > >   out:
> > > >   	rcu_read_unlock();
> > > >   	return 0;
> > > > @@ -203,3 +211,8 @@ bool intel_cap_flts_sanity(void)
> > > >   {
> > > >   	return ecap_flts(intel_iommu_ecap_sanity);
> > > >   }
> > > > +
> > > > +bool intel_cap_slts_sanity(void)
> > > > +{
> > > > +	return ecap_slts(intel_iommu_ecap_sanity);
> > > > +}
> > > > --
> > > > 2.25.1