Re: [PATCH V4 6/9] iommufd: pfn reader for file mappings

[Jason Gunthorpe <jgg@nvidia.com>]

On Fri, Oct 18, 2024 at 02:27:33PM -0700, Steve Sistare wrote:

> +static long pin_memfd_pages(struct pfn_reader_user *user, unsigned long start,
> +			    unsigned long npages)
> +{
> +	unsigned long i;
> +	unsigned long j;
> +	unsigned long offset;
> +	unsigned long npages_out = 0;
> +	struct page **upages = user->upages;
> +	unsigned long end = start + (npages << PAGE_SHIFT) - 1;
> +	long nfolios = user->ufolios_len / sizeof(*user->ufolios);
> +
> +	nfolios = memfd_pin_folios(user->file, start, end, user->ufolios,
> +				   nfolios, &offset);
> +	if (nfolios <= 0)
> +		return nfolios;
> +
> +	offset >>= PAGE_SHIFT;
> +	user->ufolios_next = user->ufolios;
> +	user->ufolios_offset = offset;
> +
> +	for (i = 0; i < nfolios; i++) {
> +		struct folio *folio = user->ufolios[i];
> +		unsigned long nr = folio_nr_pages(folio);
> +		unsigned long npin = min(nr - offset, npages);

Let's add a comment note about the loop computing npages being able to be
optimized someday

But I think this looks good, with the add_refs change you noted

Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>

> +		if (upages)
> +			for (j = offset; j < offset + npin; j++)
> +				*upages++ = folio_page(folio, j);

The upages path is 'slow' and rarely used because it is for mdevs
only, so it would be fine if it was the only loop here.

Jason