Re: [RFC PATCH v3 08/18] hw/arm/smmu-common: Add support for nested TLB

[Mostafa Saleh]

Hi Eric,

On Mon, May 20, 2024 at 10:20:43AM +0200, Eric Auger wrote:
> Hi Mostafa,
> On 5/16/24 17:20, Mostafa Saleh wrote:
> > Hi Eric,
> >
> > On Wed, May 15, 2024 at 03:48:05PM +0200, Eric Auger wrote:
> >> Hi Mostafa,
> >>
> >> On 4/29/24 05:23, Mostafa Saleh wrote:
> >>> This patch adds support for nested(combined) TLB entries.
> >> space between nested and (.
> > Will do.
> >>> The main function combine_tlb() is not used here but in the next
> >>> patches, but to simplify the patches it is introduced first.
> >>>
> >>> Main changes:
> >>> 1) New entry added in the TLB, parent_perm, for nested TLB, holds the
> >> s/entry/field, s/TLB/SMMUTLBEntry struct
> > Will do.
> >>>    stage-2 permission, this can be used to know the origin of a
> >>>    permission fault from a cached entry as caching the “and” of the
> >>>    permissions loses this information.
> >>>
> >>>    SMMUPTWEventInfo is used to hold information about PTW faults so
> >>>    the event can be populated, the value of stage (which maps to S2
> >>>    in the event) used to be set based on the current stage for TLB
> >> I don't understand "(which maps to S2 in the event)". What do you mean?
> >> This could be S1 or S2 depending on the active stage, no?
> > Not really, if the IPA size is larger than S2 input size, this is
> > considered stage-1 fault.
> >
> > For TLB permission fault, yes, that is how it is decided.
> > However, with nesting, a permission fault from a cached entry can be
> > from a stage-1 or stage-2, that’s why we now cache both and not just
> > the combined permission, and the logic to set fault stage is modified
> > accordingly.
> I meant in smmu_translate() we initially had for permission fault
> info->stage = cfg->stage whcih can be S1 or S2. Hence the fact I do not
> understand the sentence
> 
> the value of stage (which maps to S2 in the event)
> 
> I understand that with nested this computation needs to change because the 
> permission can be linked to either the S1 or S2 stage.
> Maybe that's just a matter or rephrasing?
> 

I see, that’s already how it is used now, I will rephrase it in case
it is confusing.

> 
> >>>    permission faults, however with the parent_perm, it is now set
> >>>    based on which perm has the missing permission
> >>>
> >>>    When nesting is not enabled it has the same value as perm which
> >>>    doesn't change the logic.
> >>>
> >>> 2) As combined TLB implementation is used, the combination logic
> >>>    chooses:
> >>>    - tg and level from the entry which has the smallest addr_mask.
> >> tbh I am scared bout swapping s1/s2 tg and level. In smmu_iotlb_lookup()
> >> I see tt->granule_sz being used which is s1 data. I mean it is not
> >> obvious to me this is correct. Could you maybe give more explanations
> >> detailing why/how this is guaranted to work.
> > As you mentioned the next patch reworks the lookup logic, I can reorder
> > the 2 patches if that is better, please let me know what you think?
> Yes if you manage to reorder that may be more logical because otherwise
> it looks incorrect.

Will do.
> >
> >> Can you give additional details about what s1+s2 combinations were tested?
> > I tested with S1 and S2 4K pages
> > S1 level = 3 and S2 level = 3
> > S1 level = 2 and S2 level = 3
> > S1 level = 3 and S2 level = 2
> > S1 level = 1 and S2 level = 2
> >
> > And also tested with with S1 64K granule and S2 4K.
> OK, I would suggest you mention that in the coverletter because it is
> reassuring and the combination is not totally obvious - at least to me ;-) -

Will do.

Thanks,
Mostafa
> 
> Eric
> >
> >>>    - Based on that the iova that would be cached is recalculated.
> >>>    - Translated_addr is chosen from stage-2.
> >>>
> >>> Signed-off-by: Mostafa Saleh <smostafa@google.com>
> >>> ---
> >>>  hw/arm/smmu-common.c         | 32 ++++++++++++++++++++++++++++----
> >>>  include/hw/arm/smmu-common.h |  1 +
> >>>  2 files changed, 29 insertions(+), 4 deletions(-)
> >>>
> >>> diff --git a/hw/arm/smmu-common.c b/hw/arm/smmu-common.c
> >>> index 21982621c0..0d6945fa54 100644
> >>> --- a/hw/arm/smmu-common.c
> >>> +++ b/hw/arm/smmu-common.c
> >>> @@ -394,7 +394,7 @@ static int smmu_ptw_64_s1(SMMUTransCfg *cfg,
> >>>          tlbe->entry.translated_addr = gpa;
> >>>          tlbe->entry.iova = iova & ~mask;
> >>>          tlbe->entry.addr_mask = mask;
> >>> -        tlbe->entry.perm = PTE_AP_TO_PERM(ap);
> >>> +        tlbe->parent_perm = tlbe->entry.perm = PTE_AP_TO_PERM(ap);
> >> nit: I would prefer on separate lines.
> > Will do.
> >
> >>>          tlbe->level = level;
> >>>          tlbe->granule = granule_sz;
> >>>          return 0;
> >>> @@ -515,7 +515,7 @@ static int smmu_ptw_64_s2(SMMUTransCfg *cfg,
> >>>          tlbe->entry.translated_addr = gpa;
> >>>          tlbe->entry.iova = ipa & ~mask;
> >>>          tlbe->entry.addr_mask = mask;
> >>> -        tlbe->entry.perm = s2ap;
> >>> +        tlbe->parent_perm = tlbe->entry.perm = s2ap;
> >>>          tlbe->level = level;
> >>>          tlbe->granule = granule_sz;
> >>>          return 0;
> >>> @@ -530,6 +530,27 @@ error:
> >>>      return -EINVAL;
> >>>  }
> >>>  
> >>> +/* combine 2 TLB entries and return in tlbe in nested config. */
> >> suggestion: combine S1 and S2 TLB entries into a single entry. As a
> >> result the S1 entry is overriden with combined data.
> > Will do.
> >
> >>> +static void __attribute__((unused)) combine_tlb(SMMUTLBEntry *tlbe,
> >>> +                                                SMMUTLBEntry *tlbe_s2,
> >>> +                                                dma_addr_t iova,
> >>> +                                                SMMUTransCfg *cfg)
> >>> +{
> >>> +    if (tlbe_s2->entry.addr_mask < tlbe->entry.addr_mask) {
> >>> +        tlbe->entry.addr_mask = tlbe_s2->entry.addr_mask;
> >>> +        tlbe->granule = tlbe_s2->granule;
> >>> +        tlbe->level = tlbe_s2->level;
> >>> +    }
> >>> +
> >>> +    tlbe->entry.translated_addr = CACHED_ENTRY_TO_ADDR(tlbe_s2,
> >>> +                                    tlbe->entry.translated_addr);
> >>> +
> >>> +    tlbe->entry.iova = iova & ~tlbe->entry.addr_mask;
> >>> +    /* parent_perm has s2 perm while perm has s1 perm. */
> >>  suggestion: while perm keeps s1 perm.
> >>
> > Will do.
> >
> > Thanks,
> > Mostafa
> >>> +    tlbe->parent_perm = tlbe_s2->entry.perm;
> >>> +    return;
> >>> +}
> >>> +
> >>>  /**
> >>>   * smmu_ptw - Walk the page tables for an IOVA, according to @cfg
> >>>   *
> >>> @@ -607,9 +628,12 @@ SMMUTLBEntry *smmu_translate(SMMUState *bs, 
> >>> SMMUTransCfg *cfg, dma_addr_t addr,
> >>>  
> >>>      cached_entry = smmu_iotlb_lookup(bs, cfg, &tt_combined, 
> >>> aligned_addr);
> >>>      if (cached_entry) {
> >>> -        if ((flag & IOMMU_WO) && !(cached_entry->entry.perm & IOMMU_WO)) 
> >>> {
> >>> +        if ((flag & IOMMU_WO) && !(cached_entry->entry.perm &
> >>> +            cached_entry->parent_perm & IOMMU_WO)) {
> >>>              info->type = SMMU_PTW_ERR_PERMISSION;
> >>> -            info->stage = cfg->stage;
> >>> +            info->stage = !(cached_entry->entry.perm & IOMMU_WO) ?
> >>> +                          SMMU_STAGE_1 :
> >>> +                          SMMU_STAGE_2;
> >>>              return NULL;
> >>>          }
> >>>          return cached_entry;
> >>> diff --git a/include/hw/arm/smmu-common.h b/include/hw/arm/smmu-common.h
> >>> index 09d3b9e734..1db566d451 100644
> >>> --- a/include/hw/arm/smmu-common.h
> >>> +++ b/include/hw/arm/smmu-common.h
> >>> @@ -77,6 +77,7 @@ typedef struct SMMUTLBEntry {
> >>>      IOMMUTLBEntry entry;
> >>>      uint8_t level;
> >>>      uint8_t granule;
> >>> +    IOMMUAccessFlags parent_perm;
> >>>  } SMMUTLBEntry;
> >>>  
> >>>  /* Stage-2 configuration. */
> >> Thanks
> >>
> >> Eric
> >>
>