verbs.c

 * Initialize buffer memory
 */
int
rpcrdma_buffer_create(struct rpcrdma_buffer *buf, struct rpcrdma_ep *ep,
	struct rpcrdma_ia *ia, struct rpcrdma_create_data_internal *cdata)
{
	char *p;
	size_t len;
	int i, rc;
	struct rpcrdma_mw *r;

	buf->rb_max_requests = cdata->max_requests;
	spin_lock_init(&buf->rb_lock);
	atomic_set(&buf->rb_credits, 1);

	/* Need to allocate:
	 *   1.  arrays for send and recv pointers
	 *   2.  arrays of struct rpcrdma_req to fill in pointers
	 *   3.  array of struct rpcrdma_rep for replies
	 *   4.  padding, if any
	 *   5.  mw's, fmr's or frmr's, if any
	 * Send/recv buffers in req/rep need to be registered
	 */

	len = buf->rb_max_requests *
		(sizeof(struct rpcrdma_req *) + sizeof(struct rpcrdma_rep *));
	len += cdata->padding;
	switch (ia->ri_memreg_strategy) {
	case RPCRDMA_FRMR:
		len += buf->rb_max_requests * RPCRDMA_MAX_SEGS *
				sizeof(struct rpcrdma_mw);
		break;
	case RPCRDMA_MTHCAFMR:
		/* TBD we are perhaps overallocating here */
		len += (buf->rb_max_requests + 1) * RPCRDMA_MAX_SEGS *
				sizeof(struct rpcrdma_mw);
		break;
	default:
		break;
	}

	/* allocate 1, 4 and 5 in one shot */
	p = kzalloc(len, GFP_KERNEL);
	if (p == NULL) {
		dprintk("RPC:       %s: req_t/rep_t/pad kzalloc(%zd) failed\n",
			__func__, len);
		rc = -ENOMEM;
		goto out;
	}
	buf->rb_pool = p;	/* for freeing it later */

	buf->rb_send_bufs = (struct rpcrdma_req **) p;
	p = (char *) &buf->rb_send_bufs[buf->rb_max_requests];
	buf->rb_recv_bufs = (struct rpcrdma_rep **) p;
	p = (char *) &buf->rb_recv_bufs[buf->rb_max_requests];

	/*
	 * Register the zeroed pad buffer, if any.
	 */
	if (cdata->padding) {
		rc = rpcrdma_register_internal(ia, p, cdata->padding,
					    &ep->rep_pad_mr, &ep->rep_pad);
		if (rc)
			goto out;
	}
	p += cdata->padding;

	INIT_LIST_HEAD(&buf->rb_mws);
	r = (struct rpcrdma_mw *)p;
	switch (ia->ri_memreg_strategy) {
	case RPCRDMA_FRMR:
		for (i = buf->rb_max_requests * RPCRDMA_MAX_SEGS; i; i--) {
			r->r.frmr.fr_mr = ib_alloc_fast_reg_mr(ia->ri_pd,
						ia->ri_max_frmr_depth);
			if (IS_ERR(r->r.frmr.fr_mr)) {
				rc = PTR_ERR(r->r.frmr.fr_mr);
				dprintk("RPC:       %s: ib_alloc_fast_reg_mr"
					" failed %i\n", __func__, rc);
				goto out;
			}
			r->r.frmr.fr_pgl = ib_alloc_fast_reg_page_list(
						ia->ri_id->device,
						ia->ri_max_frmr_depth);
			if (IS_ERR(r->r.frmr.fr_pgl)) {
				rc = PTR_ERR(r->r.frmr.fr_pgl);
				dprintk("RPC:       %s: "
					"ib_alloc_fast_reg_page_list "
					"failed %i\n", __func__, rc);

				ib_dereg_mr(r->r.frmr.fr_mr);
				goto out;
			}
			list_add(&r->mw_list, &buf->rb_mws);
			++r;
		}
		break;
	case RPCRDMA_MTHCAFMR:
		/* TBD we are perhaps overallocating here */
		for (i = (buf->rb_max_requests+1) * RPCRDMA_MAX_SEGS; i; i--) {
			static struct ib_fmr_attr fa =
				{ RPCRDMA_MAX_DATA_SEGS, 1, PAGE_SHIFT };
			r->r.fmr = ib_alloc_fmr(ia->ri_pd,
				IB_ACCESS_REMOTE_WRITE | IB_ACCESS_REMOTE_READ,
				&fa);
			if (IS_ERR(r->r.fmr)) {
				rc = PTR_ERR(r->r.fmr);
				dprintk("RPC:       %s: ib_alloc_fmr"
					" failed %i\n", __func__, rc);
				goto out;
			}
			list_add(&r->mw_list, &buf->rb_mws);
			++r;
		}
		break;
	default:
		break;
	}

	/*
	 * Allocate/init the request/reply buffers. Doing this
	 * using kmalloc for now -- one for each buf.
	 */
	for (i = 0; i < buf->rb_max_requests; i++) {
		struct rpcrdma_req *req;
		struct rpcrdma_rep *rep;

		len = cdata->inline_wsize + sizeof(struct rpcrdma_req);
		/* RPC layer requests *double* size + 1K RPC_SLACK_SPACE! */
		/* Typical ~2400b, so rounding up saves work later */
		if (len < 4096)
			len = 4096;
		req = kmalloc(len, GFP_KERNEL);
		if (req == NULL) {
			dprintk("RPC:       %s: request buffer %d alloc"
				" failed\n", __func__, i);
			rc = -ENOMEM;
			goto out;
		}
		memset(req, 0, sizeof(struct rpcrdma_req));
		buf->rb_send_bufs[i] = req;
		buf->rb_send_bufs[i]->rl_buffer = buf;

		rc = rpcrdma_register_internal(ia, req->rl_base,
				len - offsetof(struct rpcrdma_req, rl_base),
				&buf->rb_send_bufs[i]->rl_handle,
				&buf->rb_send_bufs[i]->rl_iov);
		if (rc)
			goto out;

		buf->rb_send_bufs[i]->rl_size = len-sizeof(struct rpcrdma_req);

		len = cdata->inline_rsize + sizeof(struct rpcrdma_rep);
		rep = kmalloc(len, GFP_KERNEL);
		if (rep == NULL) {
			dprintk("RPC:       %s: reply buffer %d alloc failed\n",
				__func__, i);
			rc = -ENOMEM;
			goto out;
		}
		memset(rep, 0, sizeof(struct rpcrdma_rep));
		buf->rb_recv_bufs[i] = rep;
		buf->rb_recv_bufs[i]->rr_buffer = buf;

		rc = rpcrdma_register_internal(ia, rep->rr_base,
				len - offsetof(struct rpcrdma_rep, rr_base),
				&buf->rb_recv_bufs[i]->rr_handle,
				&buf->rb_recv_bufs[i]->rr_iov);
		if (rc)
			goto out;

	}
	dprintk("RPC:       %s: max_requests %d\n",
		__func__, buf->rb_max_requests);
	/* done */
	return 0;
out:
	rpcrdma_buffer_destroy(buf);
	return rc;
}

/*
 * Unregister and destroy buffer memory. Need to deal with
 * partial initialization, so it's callable from failed create.
 * Must be called before destroying endpoint, as registrations
 * reference it.
 */
void
rpcrdma_buffer_destroy(struct rpcrdma_buffer *buf)
{
	int rc, i;
	struct rpcrdma_ia *ia = rdmab_to_ia(buf);
	struct rpcrdma_mw *r;

	/* clean up in reverse order from create
	 *   1.  recv mr memory (mr free, then kfree)
	 *   2.  send mr memory (mr free, then kfree)
	 *   3.  padding (if any) [moved to rpcrdma_ep_destroy]
	 *   4.  arrays
	 */
	dprintk("RPC:       %s: entering\n", __func__);

	for (i = 0; i < buf->rb_max_requests; i++) {
		if (buf->rb_recv_bufs && buf->rb_recv_bufs[i]) {
			rpcrdma_deregister_internal(ia,
					buf->rb_recv_bufs[i]->rr_handle,
					&buf->rb_recv_bufs[i]->rr_iov);
			kfree(buf->rb_recv_bufs[i]);
		}
		if (buf->rb_send_bufs && buf->rb_send_bufs[i]) {
			rpcrdma_deregister_internal(ia,
					buf->rb_send_bufs[i]->rl_handle,
					&buf->rb_send_bufs[i]->rl_iov);
			kfree(buf->rb_send_bufs[i]);
		}
	}

	while (!list_empty(&buf->rb_mws)) {
		r = list_entry(buf->rb_mws.next,
			struct rpcrdma_mw, mw_list);
		list_del(&r->mw_list);
		switch (ia->ri_memreg_strategy) {
		case RPCRDMA_FRMR:
			rc = ib_dereg_mr(r->r.frmr.fr_mr);
			if (rc)
				dprintk("RPC:       %s:"
					" ib_dereg_mr"
					" failed %i\n",
					__func__, rc);
			ib_free_fast_reg_page_list(r->r.frmr.fr_pgl);
			break;
		case RPCRDMA_MTHCAFMR:
			rc = ib_dealloc_fmr(r->r.fmr);
			if (rc)
				dprintk("RPC:       %s:"
					" ib_dealloc_fmr"
					" failed %i\n",
					__func__, rc);
			break;
		default:
			break;
		}
	}

	kfree(buf->rb_pool);
}

/*
 * Get a set of request/reply buffers.
 *
 * Reply buffer (if needed) is attached to send buffer upon return.
 * Rule:
 *    rb_send_index and rb_recv_index MUST always be pointing to the
 *    *next* available buffer (non-NULL). They are incremented after
 *    removing buffers, and decremented *before* returning them.
 */
struct rpcrdma_req *
rpcrdma_buffer_get(struct rpcrdma_buffer *buffers)
{
	struct rpcrdma_req *req;
	unsigned long flags;
	int i;
	struct rpcrdma_mw *r;

	spin_lock_irqsave(&buffers->rb_lock, flags);
	if (buffers->rb_send_index == buffers->rb_max_requests) {
		spin_unlock_irqrestore(&buffers->rb_lock, flags);
		dprintk("RPC:       %s: out of request buffers\n", __func__);
		return ((struct rpcrdma_req *)NULL);
	}

	req = buffers->rb_send_bufs[buffers->rb_send_index];
	if (buffers->rb_send_index < buffers->rb_recv_index) {
		dprintk("RPC:       %s: %d extra receives outstanding (ok)\n",
			__func__,
			buffers->rb_recv_index - buffers->rb_send_index);
		req->rl_reply = NULL;
	} else {
		req->rl_reply = buffers->rb_recv_bufs[buffers->rb_recv_index];
		buffers->rb_recv_bufs[buffers->rb_recv_index++] = NULL;
	}
	buffers->rb_send_bufs[buffers->rb_send_index++] = NULL;
	if (!list_empty(&buffers->rb_mws)) {
		i = RPCRDMA_MAX_SEGS - 1;
		do {
			r = list_entry(buffers->rb_mws.next,
					struct rpcrdma_mw, mw_list);
			list_del(&r->mw_list);
			req->rl_segments[i].mr_chunk.rl_mw = r;
		} while (--i >= 0);
	}
	spin_unlock_irqrestore(&buffers->rb_lock, flags);
	return req;
}

/*
 * Put request/reply buffers back into pool.
 * Pre-decrement counter/array index.
 */
void
rpcrdma_buffer_put(struct rpcrdma_req *req)
{
	struct rpcrdma_buffer *buffers = req->rl_buffer;
	struct rpcrdma_ia *ia = rdmab_to_ia(buffers);
	int i;
	unsigned long flags;

	BUG_ON(req->rl_nchunks != 0);
	spin_lock_irqsave(&buffers->rb_lock, flags);
	buffers->rb_send_bufs[--buffers->rb_send_index] = req;
	req->rl_niovs = 0;
	if (req->rl_reply) {
		buffers->rb_recv_bufs[--buffers->rb_recv_index] = req->rl_reply;
		req->rl_reply->rr_func = NULL;
		req->rl_reply = NULL;
	}
	switch (ia->ri_memreg_strategy) {
	case RPCRDMA_FRMR:
	case RPCRDMA_MTHCAFMR:
		/*
		 * Cycle mw's back in reverse order, and "spin" them.
		 * This delays and scrambles reuse as much as possible.
		 */
		i = 1;
		do {
			struct rpcrdma_mw **mw;
			mw = &req->rl_segments[i].mr_chunk.rl_mw;
			list_add_tail(&(*mw)->mw_list, &buffers->rb_mws);
			*mw = NULL;
		} while (++i < RPCRDMA_MAX_SEGS);
		list_add_tail(&req->rl_segments[0].mr_chunk.rl_mw->mw_list,
					&buffers->rb_mws);
		req->rl_segments[0].mr_chunk.rl_mw = NULL;
		break;
	default:
		break;
	}
	spin_unlock_irqrestore(&buffers->rb_lock, flags);
}

/*
 * Recover reply buffers from pool.
 * This happens when recovering from error conditions.
 * Post-increment counter/array index.
 */
void
rpcrdma_recv_buffer_get(struct rpcrdma_req *req)
{
	struct rpcrdma_buffer *buffers = req->rl_buffer;
	unsigned long flags;

	if (req->rl_iov.length == 0)	/* special case xprt_rdma_allocate() */
		buffers = ((struct rpcrdma_req *) buffers)->rl_buffer;
	spin_lock_irqsave(&buffers->rb_lock, flags);
	if (buffers->rb_recv_index < buffers->rb_max_requests) {
		req->rl_reply = buffers->rb_recv_bufs[buffers->rb_recv_index];
		buffers->rb_recv_bufs[buffers->rb_recv_index++] = NULL;
	}
	spin_unlock_irqrestore(&buffers->rb_lock, flags);
}

/*
 * Put reply buffers back into pool when not attached to
 * request. This happens in error conditions.
 */
void
rpcrdma_recv_buffer_put(struct rpcrdma_rep *rep)
{
	struct rpcrdma_buffer *buffers = rep->rr_buffer;
	unsigned long flags;

	rep->rr_func = NULL;
	spin_lock_irqsave(&buffers->rb_lock, flags);
	buffers->rb_recv_bufs[--buffers->rb_recv_index] = rep;
	spin_unlock_irqrestore(&buffers->rb_lock, flags);
}

/*
 * Wrappers for internal-use kmalloc memory registration, used by buffer code.
 */

int
rpcrdma_register_internal(struct rpcrdma_ia *ia, void *va, int len,
				struct ib_mr **mrp, struct ib_sge *iov)
{
	struct ib_phys_buf ipb;
	struct ib_mr *mr;
	int rc;

	/*
	 * All memory passed here was kmalloc'ed, therefore phys-contiguous.
	 */
	iov->addr = ib_dma_map_single(ia->ri_id->device,
			va, len, DMA_BIDIRECTIONAL);
	iov->length = len;

	if (ia->ri_have_dma_lkey) {
		*mrp = NULL;
		iov->lkey = ia->ri_dma_lkey;
		return 0;
	} else if (ia->ri_bind_mem != NULL) {
		*mrp = NULL;
		iov->lkey = ia->ri_bind_mem->lkey;
		return 0;
	}

	ipb.addr = iov->addr;
	ipb.size = iov->length;
	mr = ib_reg_phys_mr(ia->ri_pd, &ipb, 1,
			IB_ACCESS_LOCAL_WRITE, &iov->addr);

	dprintk("RPC:       %s: phys convert: 0x%llx "
			"registered 0x%llx length %d\n",
			__func__, (unsigned long long)ipb.addr,
			(unsigned long long)iov->addr, len);

	if (IS_ERR(mr)) {
		*mrp = NULL;
		rc = PTR_ERR(mr);
		dprintk("RPC:       %s: failed with %i\n", __func__, rc);
	} else {
		*mrp = mr;
		iov->lkey = mr->lkey;
		rc = 0;
	}

	return rc;
}

int
rpcrdma_deregister_internal(struct rpcrdma_ia *ia,
				struct ib_mr *mr, struct ib_sge *iov)
{
	int rc;

	ib_dma_unmap_single(ia->ri_id->device,
			iov->addr, iov->length, DMA_BIDIRECTIONAL);

	if (NULL == mr)
		return 0;

	rc = ib_dereg_mr(mr);
	if (rc)
		dprintk("RPC:       %s: ib_dereg_mr failed %i\n", __func__, rc);
	return rc;
}

/*
 * Wrappers for chunk registration, shared by read/write chunk code.
 */

static void
rpcrdma_map_one(struct rpcrdma_ia *ia, struct rpcrdma_mr_seg *seg, int writing)
{
	seg->mr_dir = writing ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
	seg->mr_dmalen = seg->mr_len;
	if (seg->mr_page)
		seg->mr_dma = ib_dma_map_page(ia->ri_id->device,
				seg->mr_page, offset_in_page(seg->mr_offset),
				seg->mr_dmalen, seg->mr_dir);
	else
		seg->mr_dma = ib_dma_map_single(ia->ri_id->device,
				seg->mr_offset,
				seg->mr_dmalen, seg->mr_dir);
	if (ib_dma_mapping_error(ia->ri_id->device, seg->mr_dma)) {
		dprintk("RPC:       %s: mr_dma %llx mr_offset %p mr_dma_len %zu\n",
			__func__,
			(unsigned long long)seg->mr_dma,
			seg->mr_offset, seg->mr_dmalen);
	}
}

static void
rpcrdma_unmap_one(struct rpcrdma_ia *ia, struct rpcrdma_mr_seg *seg)
{
	if (seg->mr_page)
		ib_dma_unmap_page(ia->ri_id->device,
				seg->mr_dma, seg->mr_dmalen, seg->mr_dir);
	else
		ib_dma_unmap_single(ia->ri_id->device,
				seg->mr_dma, seg->mr_dmalen, seg->mr_dir);
}

static int
rpcrdma_register_frmr_external(struct rpcrdma_mr_seg *seg,
			int *nsegs, int writing, struct rpcrdma_ia *ia,
			struct rpcrdma_xprt *r_xprt)
{
	struct rpcrdma_mr_seg *seg1 = seg;
	struct ib_send_wr invalidate_wr, frmr_wr, *bad_wr, *post_wr;

	u8 key;
	int len, pageoff;
	int i, rc;
	int seg_len;
	u64 pa;
	int page_no;

	pageoff = offset_in_page(seg1->mr_offset);
	seg1->mr_offset -= pageoff;	/* start of page */
	seg1->mr_len += pageoff;
	len = -pageoff;
	if (*nsegs > ia->ri_max_frmr_depth)
		*nsegs = ia->ri_max_frmr_depth;
	for (page_no = i = 0; i < *nsegs;) {
		rpcrdma_map_one(ia, seg, writing);
		pa = seg->mr_dma;
		for (seg_len = seg->mr_len; seg_len > 0; seg_len -= PAGE_SIZE) {
			seg1->mr_chunk.rl_mw->r.frmr.fr_pgl->
				page_list[page_no++] = pa;
			pa += PAGE_SIZE;
		}
		len += seg->mr_len;
		++seg;
		++i;
		/* Check for holes */
		if ((i < *nsegs && offset_in_page(seg->mr_offset)) ||
		    offset_in_page((seg-1)->mr_offset + (seg-1)->mr_len))
			break;
	}
	dprintk("RPC:       %s: Using frmr %p to map %d segments\n",
		__func__, seg1->mr_chunk.rl_mw, i);

	if (unlikely(seg1->mr_chunk.rl_mw->r.frmr.state == FRMR_IS_VALID)) {
		dprintk("RPC:       %s: frmr %x left valid, posting invalidate.\n",
			__func__,
			seg1->mr_chunk.rl_mw->r.frmr.fr_mr->rkey);
		/* Invalidate before using. */
		memset(&invalidate_wr, 0, sizeof invalidate_wr);
		invalidate_wr.wr_id = (unsigned long)(void *)seg1->mr_chunk.rl_mw;
		invalidate_wr.next = &frmr_wr;
		invalidate_wr.opcode = IB_WR_LOCAL_INV;
		invalidate_wr.send_flags = IB_SEND_SIGNALED;
		invalidate_wr.ex.invalidate_rkey =
			seg1->mr_chunk.rl_mw->r.frmr.fr_mr->rkey;
		DECR_CQCOUNT(&r_xprt->rx_ep);
		post_wr = &invalidate_wr;
	} else
		post_wr = &frmr_wr;

	/* Bump the key */
	key = (u8)(seg1->mr_chunk.rl_mw->r.frmr.fr_mr->rkey & 0x000000FF);
	ib_update_fast_reg_key(seg1->mr_chunk.rl_mw->r.frmr.fr_mr, ++key);

	/* Prepare FRMR WR */
	memset(&frmr_wr, 0, sizeof frmr_wr);
	frmr_wr.wr_id = (unsigned long)(void *)seg1->mr_chunk.rl_mw;
	frmr_wr.opcode = IB_WR_FAST_REG_MR;
	frmr_wr.send_flags = IB_SEND_SIGNALED;
	frmr_wr.wr.fast_reg.iova_start = seg1->mr_dma;
	frmr_wr.wr.fast_reg.page_list = seg1->mr_chunk.rl_mw->r.frmr.fr_pgl;
	frmr_wr.wr.fast_reg.page_list_len = page_no;
	frmr_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
	frmr_wr.wr.fast_reg.length = page_no << PAGE_SHIFT;
	BUG_ON(frmr_wr.wr.fast_reg.length < len);
	frmr_wr.wr.fast_reg.access_flags = (writing ?
				IB_ACCESS_REMOTE_WRITE | IB_ACCESS_LOCAL_WRITE :
				IB_ACCESS_REMOTE_READ);
	frmr_wr.wr.fast_reg.rkey = seg1->mr_chunk.rl_mw->r.frmr.fr_mr->rkey;
	DECR_CQCOUNT(&r_xprt->rx_ep);

	rc = ib_post_send(ia->ri_id->qp, post_wr, &bad_wr);

	if (rc) {
		dprintk("RPC:       %s: failed ib_post_send for register,"
			" status %i\n", __func__, rc);
		while (i--)
			rpcrdma_unmap_one(ia, --seg);
	} else {
		seg1->mr_rkey = seg1->mr_chunk.rl_mw->r.frmr.fr_mr->rkey;
		seg1->mr_base = seg1->mr_dma + pageoff;
		seg1->mr_nsegs = i;
		seg1->mr_len = len;
	}
	*nsegs = i;
	return rc;
}

static int
rpcrdma_deregister_frmr_external(struct rpcrdma_mr_seg *seg,
			struct rpcrdma_ia *ia, struct rpcrdma_xprt *r_xprt)
{
	struct rpcrdma_mr_seg *seg1 = seg;
	struct ib_send_wr invalidate_wr, *bad_wr;
	int rc;

	while (seg1->mr_nsegs--)
		rpcrdma_unmap_one(ia, seg++);

	memset(&invalidate_wr, 0, sizeof invalidate_wr);
	invalidate_wr.wr_id = (unsigned long)(void *)seg1->mr_chunk.rl_mw;
	invalidate_wr.opcode = IB_WR_LOCAL_INV;
	invalidate_wr.send_flags = IB_SEND_SIGNALED;
	invalidate_wr.ex.invalidate_rkey = seg1->mr_chunk.rl_mw->r.frmr.fr_mr->rkey;
	DECR_CQCOUNT(&r_xprt->rx_ep);

	rc = ib_post_send(ia->ri_id->qp, &invalidate_wr, &bad_wr);
	if (rc)
		dprintk("RPC:       %s: failed ib_post_send for invalidate,"
			" status %i\n", __func__, rc);
	return rc;
}

static int
rpcrdma_register_fmr_external(struct rpcrdma_mr_seg *seg,
			int *nsegs, int writing, struct rpcrdma_ia *ia)
{
	struct rpcrdma_mr_seg *seg1 = seg;
	u64 physaddrs[RPCRDMA_MAX_DATA_SEGS];
	int len, pageoff, i, rc;

	pageoff = offset_in_page(seg1->mr_offset);
	seg1->mr_offset -= pageoff;	/* start of page */
	seg1->mr_len += pageoff;
	len = -pageoff;
	if (*nsegs > RPCRDMA_MAX_DATA_SEGS)
		*nsegs = RPCRDMA_MAX_DATA_SEGS;
	for (i = 0; i < *nsegs;) {
		rpcrdma_map_one(ia, seg, writing);
		physaddrs[i] = seg->mr_dma;
		len += seg->mr_len;
		++seg;
		++i;
		/* Check for holes */
		if ((i < *nsegs && offset_in_page(seg->mr_offset)) ||
		    offset_in_page((seg-1)->mr_offset + (seg-1)->mr_len))
			break;
	}
	rc = ib_map_phys_fmr(seg1->mr_chunk.rl_mw->r.fmr,
				physaddrs, i, seg1->mr_dma);
	if (rc) {
		dprintk("RPC:       %s: failed ib_map_phys_fmr "
			"%u@0x%llx+%i (%d)... status %i\n", __func__,
			len, (unsigned long long)seg1->mr_dma,
			pageoff, i, rc);
		while (i--)
			rpcrdma_unmap_one(ia, --seg);
	} else {
		seg1->mr_rkey = seg1->mr_chunk.rl_mw->r.fmr->rkey;
		seg1->mr_base = seg1->mr_dma + pageoff;
		seg1->mr_nsegs = i;
		seg1->mr_len = len;
	}
	*nsegs = i;
	return rc;
}

static int
rpcrdma_deregister_fmr_external(struct rpcrdma_mr_seg *seg,
			struct rpcrdma_ia *ia)
{
	struct rpcrdma_mr_seg *seg1 = seg;
	LIST_HEAD(l);
	int rc;

	list_add(&seg1->mr_chunk.rl_mw->r.fmr->list, &l);
	rc = ib_unmap_fmr(&l);
	while (seg1->mr_nsegs--)
		rpcrdma_unmap_one(ia, seg++);
	if (rc)
		dprintk("RPC:       %s: failed ib_unmap_fmr,"
			" status %i\n", __func__, rc);
	return rc;
}

int
rpcrdma_register_external(struct rpcrdma_mr_seg *seg,
			int nsegs, int writing, struct rpcrdma_xprt *r_xprt)
{
	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
	int rc = 0;

	switch (ia->ri_memreg_strategy) {

#if RPCRDMA_PERSISTENT_REGISTRATION
	case RPCRDMA_ALLPHYSICAL:
		rpcrdma_map_one(ia, seg, writing);
		seg->mr_rkey = ia->ri_bind_mem->rkey;
		seg->mr_base = seg->mr_dma;
		seg->mr_nsegs = 1;
		nsegs = 1;
		break;
#endif

	/* Registration using frmr registration */
	case RPCRDMA_FRMR:
		rc = rpcrdma_register_frmr_external(seg, &nsegs, writing, ia, r_xprt);
		break;

	/* Registration using fmr memory registration */
	case RPCRDMA_MTHCAFMR:
		rc = rpcrdma_register_fmr_external(seg, &nsegs, writing, ia);
		break;

	default:
		return -1;
	}
	if (rc)
		return -1;

	return nsegs;
}

int
rpcrdma_deregister_external(struct rpcrdma_mr_seg *seg,
		struct rpcrdma_xprt *r_xprt)
{
	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
	int nsegs = seg->mr_nsegs, rc;

	switch (ia->ri_memreg_strategy) {

#if RPCRDMA_PERSISTENT_REGISTRATION
	case RPCRDMA_ALLPHYSICAL:
		BUG_ON(nsegs != 1);
		rpcrdma_unmap_one(ia, seg);
		rc = 0;
		break;
#endif

	case RPCRDMA_FRMR:
		rc = rpcrdma_deregister_frmr_external(seg, ia, r_xprt);
		break;

	case RPCRDMA_MTHCAFMR:
		rc = rpcrdma_deregister_fmr_external(seg, ia);
		break;

	default:
		break;
	}
	return nsegs;
}

/*
 * Prepost any receive buffer, then post send.
 *
 * Receive buffer is donated to hardware, reclaimed upon recv completion.
 */
int
rpcrdma_ep_post(struct rpcrdma_ia *ia,
		struct rpcrdma_ep *ep,
		struct rpcrdma_req *req)
{
	struct ib_send_wr send_wr, *send_wr_fail;
	struct rpcrdma_rep *rep = req->rl_reply;
	int rc;

	if (rep) {
		rc = rpcrdma_ep_post_recv(ia, ep, rep);
		if (rc)
			goto out;
		req->rl_reply = NULL;
	}

	send_wr.next = NULL;
	send_wr.wr_id = 0ULL;	/* no send cookie */
	send_wr.sg_list = req->rl_send_iov;
	send_wr.num_sge = req->rl_niovs;
	send_wr.opcode = IB_WR_SEND;
	if (send_wr.num_sge == 4)	/* no need to sync any pad (constant) */
		ib_dma_sync_single_for_device(ia->ri_id->device,
			req->rl_send_iov[3].addr, req->rl_send_iov[3].length,
			DMA_TO_DEVICE);
	ib_dma_sync_single_for_device(ia->ri_id->device,
		req->rl_send_iov[1].addr, req->rl_send_iov[1].length,
		DMA_TO_DEVICE);
	ib_dma_sync_single_for_device(ia->ri_id->device,
		req->rl_send_iov[0].addr, req->rl_send_iov[0].length,
		DMA_TO_DEVICE);

	if (DECR_CQCOUNT(ep) > 0)
		send_wr.send_flags = 0;
	else { /* Provider must take a send completion every now and then */
		INIT_CQCOUNT(ep);
		send_wr.send_flags = IB_SEND_SIGNALED;
	}

	rc = ib_post_send(ia->ri_id->qp, &send_wr, &send_wr_fail);
	if (rc)
		dprintk("RPC:       %s: ib_post_send returned %i\n", __func__,
			rc);
out:
	return rc;
}

/*
 * (Re)post a receive buffer.
 */
int
rpcrdma_ep_post_recv(struct rpcrdma_ia *ia,
		     struct rpcrdma_ep *ep,
		     struct rpcrdma_rep *rep)
{
	struct ib_recv_wr recv_wr, *recv_wr_fail;
	int rc;

	recv_wr.next = NULL;
	recv_wr.wr_id = (u64) (unsigned long) rep;
	recv_wr.sg_list = &rep->rr_iov;
	recv_wr.num_sge = 1;

	ib_dma_sync_single_for_cpu(ia->ri_id->device,
		rep->rr_iov.addr, rep->rr_iov.length, DMA_BIDIRECTIONAL);

	rc = ib_post_recv(ia->ri_id->qp, &recv_wr, &recv_wr_fail);

	if (rc)
		dprintk("RPC:       %s: ib_post_recv returned %i\n", __func__,
			rc);
	return rc;
}