On Fri, Sep 08, 2023 at 03:26:47PM +0900, Daisuke Matsuda wrote:
> +static inline bool rxe_odp_check_pages(struct rxe_mr *mr, u64 iova,
> + int length, u32 flags)
> +{
> + unsigned long lower, upper, idx;
> + unsigned long hmm_flags = HMM_PFN_VALID;
> + struct ib_umem_odp *umem_odp = to_ib_umem_odp(mr->umem);
> + struct page *page;
> + bool need_fault = false;
> +
> + lower = rxe_mr_iova_to_index(mr, iova);
> + upper = rxe_mr_iova_to_index(mr, iova + length - 1);
> +
> + if (!(flags & RXE_PAGEFAULT_RDONLY))
> + hmm_flags |= HMM_PFN_WRITE;
> +
> + /* xarray is protected by umem_mutex */
> + for (idx = lower; idx <= upper; idx++) {
> + page = xa_load(&mr->page_list, idx);
> +
> + if (!page || !(umem_odp->pfn_list[idx] & hmm_flags)) {
> + need_fault = true;
Again you don't need the pfn_list and rxe should perhaps ideally find
some way to disable it since we store struct pages in the xarray.
This could also be a xas loop
Jason
