Commit 54375e73 authored by Jason Gunthorpe's avatar Jason Gunthorpe
Browse files

RDMA/mlx5: Split implicit handling from pagefault_mr 

The single routine has a very confusing scheme to advance to the next
child MR when working on an implicit parent. This scheme can only be used
when working with an implicit parent and must not be triggered when
working on a normal MR.

Re-arrange things by directly putting all the single-MR stuff into one
function and calling it in a loop for the implicit case. Simplify some of
the error handling in the new pagefault_real_mr() to remove unneeded gotos.

Link: https://lore.kernel.org/r/20191009160934.3143-9-jgg@ziepe.ca


Reviewed-by: default avatarArtemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: default avatarJason Gunthorpe <jgg@mellanox.com>
parent 9162420d

drivers/infiniband/hw/mlx5/odp.c

+76 −49
Original line number Diff line number Diff line
@@ -629,33 +629,18 @@ void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *imr) 
}



#define MLX5_PF_FLAGS_DOWNGRADE BIT(1)

static int pagefault_mr(struct mlx5_ib_mr *mr, u64 io_virt, size_t bcnt,

			u32 *bytes_mapped, u32 flags)

static int pagefault_real_mr(struct mlx5_ib_mr *mr, struct ib_umem_odp *odp,

			     u64 user_va, size_t bcnt, u32 *bytes_mapped,

			     u32 flags)

{

	int npages = 0, current_seq, page_shift, ret, np;

	struct ib_umem_odp *odp_mr = to_ib_umem_odp(mr->umem);

	int current_seq, page_shift, ret, np;

	bool downgrade = flags & MLX5_PF_FLAGS_DOWNGRADE;

	u64 access_mask;

	u64 start_idx, page_mask;

	struct ib_umem_odp *odp;

	size_t size;



	if (odp_mr->is_implicit_odp) {

		odp = implicit_mr_get_data(mr, io_virt, bcnt);



		if (IS_ERR(odp))

			return PTR_ERR(odp);

		mr = odp->private;

	} else {

		odp = odp_mr;

	}



next_mr:

	size = min_t(size_t, bcnt, ib_umem_end(odp) - io_virt);



	page_shift = odp->page_shift;

	page_mask = ~(BIT(page_shift) - 1);

	start_idx = (io_virt - (mr->mmkey.iova & page_mask)) >> page_shift;

	start_idx = (user_va - (mr->mmkey.iova & page_mask)) >> page_shift;

	access_mask = ODP_READ_ALLOWED_BIT;



	if (odp->umem.writable && !downgrade)
@@ -668,13 +653,10 @@ next_mr: 
	 */

	smp_rmb();



	ret = ib_umem_odp_map_dma_pages(odp, io_virt, size, access_mask,

	np = ib_umem_odp_map_dma_pages(odp, user_va, bcnt, access_mask,

				       current_seq);



	if (ret < 0)

		goto out;



	np = ret;

	if (np < 0)

		return np;



	mutex_lock(&odp->umem_mutex);

	if (!ib_umem_mmu_notifier_retry(odp, current_seq)) {
@@ -699,31 +681,12 @@ next_mr: 


	if (bytes_mapped) {

		u32 new_mappings = (np << page_shift) -

			(io_virt - round_down(io_virt, 1 << page_shift));

		*bytes_mapped += min_t(u32, new_mappings, size);

	}



	npages += np << (page_shift - PAGE_SHIFT);

	bcnt -= size;



	if (unlikely(bcnt)) {

		struct ib_umem_odp *next;

			(user_va - round_down(user_va, 1 << page_shift));



		io_virt += size;

		next = odp_next(odp);

		if (unlikely(!next || ib_umem_start(next) != io_virt)) {

			mlx5_ib_dbg(

				mr->dev,

				"next implicit leaf removed at 0x%llx. got %p\n",

				io_virt, next);

			return -EAGAIN;

		}

		odp = next;

		mr = odp->private;

		goto next_mr;

		*bytes_mapped += min_t(u32, new_mappings, bcnt);

	}



	return npages;

	return np << (page_shift - PAGE_SHIFT);



out:

	if (ret == -EAGAIN) {
@@ -742,6 +705,70 @@ out: 
	return ret;

}



/*

 * Returns:

 *  -EFAULT: The io_virt->bcnt is not within the MR, it covers pages that are

 *           not accessible, or the MR is no longer valid.

 *  -EAGAIN/-ENOMEM: The operation should be retried

 *

 *  -EINVAL/others: General internal malfunction

 *  >0: Number of pages mapped

 */

static int pagefault_mr(struct mlx5_ib_mr *mr, u64 io_virt, size_t bcnt,

			u32 *bytes_mapped, u32 flags)

{

	struct ib_umem_odp *odp = to_ib_umem_odp(mr->umem);

	struct ib_umem_odp *child;

	int npages = 0;



	if (!odp->is_implicit_odp) {

		if (unlikely(io_virt < ib_umem_start(odp) ||

			     ib_umem_end(odp) - io_virt < bcnt))

			return -EFAULT;

		return pagefault_real_mr(mr, odp, io_virt, bcnt, bytes_mapped,

					 flags);

	}



	if (unlikely(io_virt >= mlx5_imr_ksm_entries * MLX5_IMR_MTT_SIZE ||

		     mlx5_imr_ksm_entries * MLX5_IMR_MTT_SIZE - io_virt < bcnt))

		return -EFAULT;



	child = implicit_mr_get_data(mr, io_virt, bcnt);

	if (IS_ERR(child))

		return PTR_ERR(child);



	/* Fault each child mr that intersects with our interval. */

	while (bcnt) {

		u64 end = min_t(u64, io_virt + bcnt, ib_umem_end(child));

		u64 len = end - io_virt;

		int ret;



		ret = pagefault_real_mr(child->private, child, io_virt, len,

					bytes_mapped, flags);

		if (ret < 0)

			return ret;

		io_virt += len;

		bcnt -= len;

		npages += ret;



		if (unlikely(bcnt)) {

			child = odp_next(child);

			/*

			 * implicit_mr_get_data sets up all the leaves, this

			 * means they got invalidated before we got to them.

			 */

			if (!child || ib_umem_start(child) != io_virt) {

				mlx5_ib_dbg(

					mr->dev,

					"next implicit leaf removed at 0x%llx.\n",

					io_virt);

				return -EAGAIN;

			}

		}

	}

	return npages;

}



struct pf_frame {

	struct pf_frame *next;

	u32 key;

CRA Git | Maintained and supported by SUSTech CRA and CCSE