Re: [PATCH V2 3/9] iommufd: pfn reader for file mappings

[Nicolin Chen <nicolinc@nvidia.com>]

On Tue, Sep 24, 2024 at 08:05:32AM -0700, Steve Sistare wrote:

> diff --git a/drivers/iommu/iommufd/pages.c b/drivers/iommu/iommufd/pages.c
> index 69822d4..df5ba4f 100644
> --- a/drivers/iommu/iommufd/pages.c
> +++ b/drivers/iommu/iommufd/pages.c
> @@ -703,19 +703,28 @@ struct pfn_reader_user {
>          * neither
>          */
>         int locked;
> +
> +       /* The following are only valid if file != NULL. */
> +       struct file *file;
> +       struct folio **ufolios;
> +       unsigned long ufolios_len;
>  };

I am not 100% sure. Yet, given the comments at pfn_reader_user:
	/* pfn_reader_user is just the pin_user_pages() path */
which FILE doesn't use, I think we might need a pfn_reader_file
with own set of helpers:
	pfn_reader_file_init
	pfn_reader_file_destroy
	pfn_reader_file_pin
Then their callers should switch-case pages->type to select the
corresponding type of pfn_reader functions.

Maybe USER and FILE don't diff that much. But considering we'll
next add a PHYS type and none of them makes sense to PHYS, there
could be another layer of cleanup at these pfn_reader functions,
v.s. stuffing all FILE-related lines into pfn_reader_user.

Thanks
Nicolin