Re: [PATCH v7 05/15] iommupt: Add iova_to_phys op

[Samiullah Khawaja <skhawaja@google.com>]

On Thu, Oct 23, 2025 at 11:21 AM Jason Gunthorpe <jgg@nvidia.com> wrote:
>
> iova_to_phys is a performance path for the DMA API and iommufd, implement
> it using an unrolled get_user_pages() like function waterfall scheme.
>
> The implementation itself is fairly trivial.
>
> Tested-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
> Reviewed-by: Kevin Tian <kevin.tian@intel.com>
> Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
> ---
>  drivers/iommu/generic_pt/iommu_pt.h | 105 ++++++++++++++++++++++++++++
>  include/linux/generic_pt/iommu.h    |  19 +++--
>  2 files changed, 119 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/iommu/generic_pt/iommu_pt.h b/drivers/iommu/generic_pt/iommu_pt.h
> index 564f2d3a6e11e1..5ff1b887928a46 100644
> --- a/drivers/iommu/generic_pt/iommu_pt.h
> +++ b/drivers/iommu/generic_pt/iommu_pt.h
> @@ -17,6 +17,111 @@
>
>  #define DOMAIN_NS(op) CONCATENATE(CONCATENATE(pt_iommu_, PTPFX), op)
>
> +static int make_range_ul(struct pt_common *common, struct pt_range *range,
> +                        unsigned long iova, unsigned long len)
> +{
> +       unsigned long last;
> +
> +       if (unlikely(len == 0))
> +               return -EINVAL;
> +
> +       if (check_add_overflow(iova, len - 1, &last))
> +               return -EOVERFLOW;
> +
> +       *range = pt_make_range(common, iova, last);
> +       if (sizeof(iova) > sizeof(range->va)) {
> +               if (unlikely(range->va != iova || range->last_va != last))
> +                       return -EOVERFLOW;
> +       }
> +       return 0;
> +}
> +
> +static __maybe_unused int make_range_u64(struct pt_common *common,
> +                                        struct pt_range *range, u64 iova,
> +                                        u64 len)
> +{
> +       if (unlikely(iova > ULONG_MAX || len > ULONG_MAX))
> +               return -EOVERFLOW;
> +       return make_range_ul(common, range, iova, len);
> +}
> +
> +/*
> + * Some APIs use unsigned long, while othersuse dma_addr_t as the type. Dispatch
> + * to the correct validation based on the type.
> + */
> +#define make_range_no_check(common, range, iova, len)                   \
> +       ({                                                              \
> +               int ret;                                                \
> +               if (sizeof(iova) > sizeof(unsigned long) ||             \
> +                   sizeof(len) > sizeof(unsigned long))                \
> +                       ret = make_range_u64(common, range, iova, len); \
> +               else                                                    \
> +                       ret = make_range_ul(common, range, iova, len);  \
> +               ret;                                                    \
> +       })
> +
> +#define make_range(common, range, iova, len)                             \
> +       ({                                                               \
> +               int ret = make_range_no_check(common, range, iova, len); \
> +               if (!ret)                                                \
> +                       ret = pt_check_range(range);                     \
> +               ret;                                                     \
> +       })
> +
> +static __always_inline int __do_iova_to_phys(struct pt_range *range, void *arg,
> +                                            unsigned int level,
> +                                            struct pt_table_p *table,
> +                                            pt_level_fn_t descend_fn)
> +{
> +       struct pt_state pts = pt_init(range, level, table);
> +       pt_oaddr_t *res = arg;
> +
> +       switch (pt_load_single_entry(&pts)) {
> +       case PT_ENTRY_EMPTY:
> +               return -ENOENT;
> +       case PT_ENTRY_TABLE:
> +               return pt_descend(&pts, arg, descend_fn);
> +       case PT_ENTRY_OA:
> +               *res = pt_entry_oa_exact(&pts);
> +               return 0;
> +       }
> +       return -ENOENT;
> +}
> +PT_MAKE_LEVELS(__iova_to_phys, __do_iova_to_phys);
> +
> +/**
> + * iova_to_phys() - Return the output address for the given IOVA
> + * @iommu_table: Table to query
> + * @iova: IO virtual address to query
> + *
> + * Determine the output address from the given IOVA. @iova may have any
> + * alignment, the returned physical will be adjusted with any sub page offset.
> + *
> + * Context: The caller must hold a read range lock that includes @iova.
> + *
> + * Return: 0 if there is no translation for the given iova.
> + */
> +phys_addr_t DOMAIN_NS(iova_to_phys)(struct iommu_domain *domain,
> +                                   dma_addr_t iova)
> +{
> +       struct pt_iommu *iommu_table =
> +               container_of(domain, struct pt_iommu, domain);
> +       struct pt_range range;
> +       pt_oaddr_t res;
> +       int ret;
> +
> +       ret = make_range(common_from_iommu(iommu_table), &range, iova, 1);
> +       if (ret)
> +               return ret;
> +
> +       ret = pt_walk_range(&range, __iova_to_phys, &res);
> +       /* PHYS_ADDR_MAX would be a better error code */
> +       if (ret)
> +               return 0;
> +       return res;
> +}
> +EXPORT_SYMBOL_NS_GPL(DOMAIN_NS(iova_to_phys), "GENERIC_PT_IOMMU");
> +
>  struct pt_iommu_collect_args {
>         struct iommu_pages_list free_list;
>  };
> diff --git a/include/linux/generic_pt/iommu.h b/include/linux/generic_pt/iommu.h
> index dc731fe003d153..5622856e199881 100644
> --- a/include/linux/generic_pt/iommu.h
> +++ b/include/linux/generic_pt/iommu.h
> @@ -116,11 +116,13 @@ struct pt_iommu_cfg {
>  };
>
>  /* Generate the exported function signatures from iommu_pt.h */
> -#define IOMMU_PROTOTYPES(fmt)                                             \
> -       int pt_iommu_##fmt##_init(struct pt_iommu_##fmt *table,           \
> -                                 const struct pt_iommu_##fmt##_cfg *cfg, \
> -                                 gfp_t gfp);                             \
> -       void pt_iommu_##fmt##_hw_info(struct pt_iommu_##fmt *table,       \
> +#define IOMMU_PROTOTYPES(fmt)                                                  \
> +       phys_addr_t pt_iommu_##fmt##_iova_to_phys(struct iommu_domain *domain, \
> +                                                 dma_addr_t iova);            \
> +       int pt_iommu_##fmt##_init(struct pt_iommu_##fmt *table,                \
> +                                 const struct pt_iommu_##fmt##_cfg *cfg,      \
> +                                 gfp_t gfp);                                  \
> +       void pt_iommu_##fmt##_hw_info(struct pt_iommu_##fmt *table,            \
>                                       struct pt_iommu_##fmt##_hw_info *info)
>  #define IOMMU_FORMAT(fmt, member)       \
>         struct pt_iommu_##fmt {         \
> @@ -129,6 +131,13 @@ struct pt_iommu_cfg {
>         };                              \
>         IOMMU_PROTOTYPES(fmt)
>
> +/*
> + * A driver uses IOMMU_PT_DOMAIN_OPS to populate the iommu_domain_ops for the
> + * iommu_pt
> + */
> +#define IOMMU_PT_DOMAIN_OPS(fmt) \
> +       .iova_to_phys = &pt_iommu_##fmt##_iova_to_phys,
> +
>  /*
>   * The driver should setup its domain struct like
>   *     union {
> --
> 2.43.0
>
>

Reviewed-by: Samiullah Khawaja <skhawaja@google.com>