[linux-2.6-block.git] / drivers / infiniband / hw / mlx4 / srq.c

/*
 * Copyright (c) 2007 Cisco Systems, Inc. All rights reserved.
 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <linux/mlx4/qp.h>
#include <linux/mlx4/srq.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>

#include "mlx4_ib.h"
#include "user.h"

static void *get_wqe(struct mlx4_ib_srq *srq, int n)
{
	return mlx4_buf_offset(&srq->buf, n << srq->msrq.wqe_shift);
}

static void mlx4_ib_srq_event(struct mlx4_srq *srq, enum mlx4_event type)
{
	struct ib_event event;
	struct ib_srq *ibsrq = &to_mibsrq(srq)->ibsrq;

	if (ibsrq->event_handler) {
		event.device      = ibsrq->device;
		event.element.srq = ibsrq;
		switch (type) {
		case MLX4_EVENT_TYPE_SRQ_LIMIT:
			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
			break;
		case MLX4_EVENT_TYPE_SRQ_CATAS_ERROR:
			event.event = IB_EVENT_SRQ_ERR;
			break;
		default:
			pr_warn("Unexpected event type %d "
			       "on SRQ %06x\n", type, srq->srqn);
			return;
		}

		ibsrq->event_handler(&event, ibsrq->srq_context);
	}
}

struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
				  struct ib_srq_init_attr *init_attr,
				  struct ib_udata *udata)
{
	struct mlx4_ib_dev *dev = to_mdev(pd->device);
	struct mlx4_ib_srq *srq;
	struct mlx4_wqe_srq_next_seg *next;
	struct mlx4_wqe_data_seg *scatter;
	u32 cqn;
	u16 xrcdn;
	int desc_size;
	int buf_size;
	int err;
	int i;

	/* Sanity check SRQ size before proceeding */
	if (init_attr->attr.max_wr  >= dev->dev->caps.max_srq_wqes ||
	    init_attr->attr.max_sge >  dev->dev->caps.max_srq_sge)
		return ERR_PTR(-EINVAL);

	srq = kmalloc(sizeof *srq, GFP_KERNEL);
	if (!srq)
		return ERR_PTR(-ENOMEM);

	mutex_init(&srq->mutex);
	spin_lock_init(&srq->lock);
	srq->msrq.max    = roundup_pow_of_two(init_attr->attr.max_wr + 1);
	srq->msrq.max_gs = init_attr->attr.max_sge;

	desc_size = max(32UL,
			roundup_pow_of_two(sizeof (struct mlx4_wqe_srq_next_seg) +
					   srq->msrq.max_gs *
					   sizeof (struct mlx4_wqe_data_seg)));
	srq->msrq.wqe_shift = ilog2(desc_size);

	buf_size = srq->msrq.max * desc_size;

	if (pd->uobject) {
		struct mlx4_ib_create_srq ucmd;

		if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
			err = -EFAULT;
			goto err_srq;
		}

		srq->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr,
					buf_size, 0, 0);
		if (IS_ERR(srq->umem)) {
			err = PTR_ERR(srq->umem);
			goto err_srq;
		}

		err = mlx4_mtt_init(dev->dev, ib_umem_page_count(srq->umem),
				    ilog2(srq->umem->page_size), &srq->mtt);
		if (err)
			goto err_buf;

		err = mlx4_ib_umem_write_mtt(dev, &srq->mtt, srq->umem);
		if (err)
			goto err_mtt;

		err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context),
					  ucmd.db_addr, &srq->db);
		if (err)
			goto err_mtt;
	} else {
		err = mlx4_db_alloc(dev->dev, &srq->db, 0, GFP_KERNEL);
		if (err)
			goto err_srq;

		*srq->db.db = 0;

		if (mlx4_buf_alloc(dev->dev, buf_size, PAGE_SIZE * 2, &srq->buf,
				   GFP_KERNEL)) {
			err = -ENOMEM;
			goto err_db;
		}

		srq->head    = 0;
		srq->tail    = srq->msrq.max - 1;
		srq->wqe_ctr = 0;

		for (i = 0; i < srq->msrq.max; ++i) {
			next = get_wqe(srq, i);
			next->next_wqe_index =
				cpu_to_be16((i + 1) & (srq->msrq.max - 1));

			for (scatter = (void *) (next + 1);
			     (void *) scatter < (void *) next + desc_size;
			     ++scatter)
				scatter->lkey = cpu_to_be32(MLX4_INVALID_LKEY);
		}

		err = mlx4_mtt_init(dev->dev, srq->buf.npages, srq->buf.page_shift,
				    &srq->mtt);
		if (err)
			goto err_buf;

		err = mlx4_buf_write_mtt(dev->dev, &srq->mtt, &srq->buf, GFP_KERNEL);
		if (err)
			goto err_mtt;

		srq->wrid = kmalloc(srq->msrq.max * sizeof (u64), GFP_KERNEL);
		if (!srq->wrid) {
			srq->wrid = __vmalloc(srq->msrq.max * sizeof(u64),
					      GFP_KERNEL, PAGE_KERNEL);
			if (!srq->wrid) {
				err = -ENOMEM;
				goto err_mtt;
			}
		}
	}

	cqn = (init_attr->srq_type == IB_SRQT_XRC) ?
		to_mcq(init_attr->ext.xrc.cq)->mcq.cqn : 0;
	xrcdn = (init_attr->srq_type == IB_SRQT_XRC) ?
		to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn :
		(u16) dev->dev->caps.reserved_xrcds;
	err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, cqn, xrcdn, &srq->mtt,
			     srq->db.dma, &srq->msrq);
	if (err)
		goto err_wrid;

	srq->msrq.event = mlx4_ib_srq_event;
	srq->ibsrq.ext.xrc.srq_num = srq->msrq.srqn;

	if (pd->uobject)
		if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof (__u32))) {
			err = -EFAULT;
			goto err_wrid;
		}

	init_attr->attr.max_wr = srq->msrq.max - 1;

	return &srq->ibsrq;

err_wrid:
	if (pd->uobject)
		mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context), &srq->db);
	else
		kvfree(srq->wrid);

err_mtt:
	mlx4_mtt_cleanup(dev->dev, &srq->mtt);

err_buf:
	if (pd->uobject)
		ib_umem_release(srq->umem);
	else
		mlx4_buf_free(dev->dev, buf_size, &srq->buf);

err_db:
	if (!pd->uobject)
		mlx4_db_free(dev->dev, &srq->db);

err_srq:
	kfree(srq);

	return ERR_PTR(err);
}

int mlx4_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata)
{
	struct mlx4_ib_dev *dev = to_mdev(ibsrq->device);
	struct mlx4_ib_srq *srq = to_msrq(ibsrq);
	int ret;

	/* We don't support resizing SRQs (yet?) */
	if (attr_mask & IB_SRQ_MAX_WR)
		return -EINVAL;

	if (attr_mask & IB_SRQ_LIMIT) {
		if (attr->srq_limit >= srq->msrq.max)
			return -EINVAL;

		mutex_lock(&srq->mutex);
		ret = mlx4_srq_arm(dev->dev, &srq->msrq, attr->srq_limit);
		mutex_unlock(&srq->mutex);

		if (ret)
			return ret;
	}

	return 0;
}

int mlx4_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
{
	struct mlx4_ib_dev *dev = to_mdev(ibsrq->device);
	struct mlx4_ib_srq *srq = to_msrq(ibsrq);
	int ret;
	int limit_watermark;

	ret = mlx4_srq_query(dev->dev, &srq->msrq, &limit_watermark);
	if (ret)
		return ret;

	srq_attr->srq_limit = limit_watermark;
	srq_attr->max_wr    = srq->msrq.max - 1;
	srq_attr->max_sge   = srq->msrq.max_gs;

	return 0;
}

int mlx4_ib_destroy_srq(struct ib_srq *srq)
{
	struct mlx4_ib_dev *dev = to_mdev(srq->device);
	struct mlx4_ib_srq *msrq = to_msrq(srq);

	mlx4_srq_free(dev->dev, &msrq->msrq);
	mlx4_mtt_cleanup(dev->dev, &msrq->mtt);

	if (srq->uobject) {
		mlx4_ib_db_unmap_user(to_mucontext(srq->uobject->context), &msrq->db);
		ib_umem_release(msrq->umem);
	} else {
		kfree(msrq->wrid);
		mlx4_buf_free(dev->dev, msrq->msrq.max << msrq->msrq.wqe_shift,
			      &msrq->buf);
		mlx4_db_free(dev->dev, &msrq->db);
	}

	kfree(msrq);

	return 0;
}

void mlx4_ib_free_srq_wqe(struct mlx4_ib_srq *srq, int wqe_index)
{
	struct mlx4_wqe_srq_next_seg *next;

	/* always called with interrupts disabled. */
	spin_lock(&srq->lock);

	next = get_wqe(srq, srq->tail);
	next->next_wqe_index = cpu_to_be16(wqe_index);
	srq->tail = wqe_index;

	spin_unlock(&srq->lock);
}

int mlx4_ib_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
			  struct ib_recv_wr **bad_wr)
{
	struct mlx4_ib_srq *srq = to_msrq(ibsrq);
	struct mlx4_wqe_srq_next_seg *next;
	struct mlx4_wqe_data_seg *scat;
	unsigned long flags;
	int err = 0;
	int nreq;
	int i;
	struct mlx4_ib_dev *mdev = to_mdev(ibsrq->device);

	spin_lock_irqsave(&srq->lock, flags);
	if (mdev->dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
		err = -EIO;
		*bad_wr = wr;
		nreq = 0;
		goto out;
	}

	for (nreq = 0; wr; ++nreq, wr = wr->next) {
		if (unlikely(wr->num_sge > srq->msrq.max_gs)) {
			err = -EINVAL;
			*bad_wr = wr;
			break;
		}

		if (unlikely(srq->head == srq->tail)) {
			err = -ENOMEM;
			*bad_wr = wr;
			break;
		}

		srq->wrid[srq->head] = wr->wr_id;

		next      = get_wqe(srq, srq->head);
		srq->head = be16_to_cpu(next->next_wqe_index);
		scat      = (struct mlx4_wqe_data_seg *) (next + 1);

		for (i = 0; i < wr->num_sge; ++i) {
			scat[i].byte_count = cpu_to_be32(wr->sg_list[i].length);
			scat[i].lkey       = cpu_to_be32(wr->sg_list[i].lkey);
			scat[i].addr       = cpu_to_be64(wr->sg_list[i].addr);
		}

		if (i < srq->msrq.max_gs) {
			scat[i].byte_count = 0;
			scat[i].lkey       = cpu_to_be32(MLX4_INVALID_LKEY);
			scat[i].addr       = 0;
		}
	}

	if (likely(nreq)) {
		srq->wqe_ctr += nreq;

		/*
		 * Make sure that descriptors are written before
		 * doorbell record.
		 */
		wmb();

		*srq->db.db = cpu_to_be32(srq->wqe_ctr);
	}
out:

	spin_unlock_irqrestore(&srq->lock, flags);

	return err;
}
Commit	Line	Data
225c7b1f RD	1	/*
225c7b1f RD	2	* Copyright (c) 2007 Cisco Systems, Inc. All rights reserved.
51a379d0	3	* Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
225c7b1f RD	4	*
	5	* This software is available to you under a choice of one of two
	6	* licenses. You may choose to be licensed under the terms of the GNU
	7	* General Public License (GPL) Version 2, available from the file
	8	* COPYING in the main directory of this source tree, or the
	9	* OpenIB.org BSD license below:
	10	*
	11	* Redistribution and use in source and binary forms, with or
	12	* without modification, are permitted provided that the following
	13	* conditions are met:
	14	*
	15	* - Redistributions of source code must retain the above
	16	* copyright notice, this list of conditions and the following
	17	* disclaimer.
	18	*
	19	* - Redistributions in binary form must reproduce the above
	20	* copyright notice, this list of conditions and the following
	21	* disclaimer in the documentation and/or other materials
	22	* provided with the distribution.
	23	*
	24	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
	25	* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
	26	* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
	27	* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
	28	* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
	29	* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
	30	* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
	31	* SOFTWARE.
	32	*/
	33
	34	#include <linux/mlx4/qp.h>
	35	#include <linux/mlx4/srq.h>
5a0e3ad6	36	#include <linux/slab.h>
0ef2f05c	37	#include <linux/vmalloc.h>
225c7b1f RD	38
	39	#include "mlx4_ib.h"
	40	#include "user.h"
	41
	42	static void get_wqe(struct mlx4_ib_srq srq, int n)
	43	{
1c69fc2a	44	return mlx4_buf_offset(&srq->buf, n << srq->msrq.wqe_shift);
225c7b1f RD	45	}
	46
	47	static void mlx4_ib_srq_event(struct mlx4_srq *srq, enum mlx4_event type)
	48	{
	49	struct ib_event event;
	50	struct ib_srq *ibsrq = &to_mibsrq(srq)->ibsrq;
	51
	52	if (ibsrq->event_handler) {
	53	event.device = ibsrq->device;
	54	event.element.srq = ibsrq;
	55	switch (type) {
	56	case MLX4_EVENT_TYPE_SRQ_LIMIT:
	57	event.event = IB_EVENT_SRQ_LIMIT_REACHED;
	58	break;
	59	case MLX4_EVENT_TYPE_SRQ_CATAS_ERROR:
	60	event.event = IB_EVENT_SRQ_ERR;
	61	break;
	62	default:
987c8f8f	63	pr_warn("Unexpected event type %d "
225c7b1f RD	64	"on SRQ %06x\n", type, srq->srqn);
	65	return;
	66	}
	67
	68	ibsrq->event_handler(&event, ibsrq->srq_context);
	69	}
	70	}
	71
	72	struct ib_srq mlx4_ib_create_srq(struct ib_pd pd,
	73	struct ib_srq_init_attr *init_attr,
	74	struct ib_udata *udata)
	75	{
	76	struct mlx4_ib_dev *dev = to_mdev(pd->device);
	77	struct mlx4_ib_srq *srq;
	78	struct mlx4_wqe_srq_next_seg *next;
4c425588	79	struct mlx4_wqe_data_seg *scatter;
18abd5ea SH	80	u32 cqn;
18abd5ea SH	81	u16 xrcdn;
225c7b1f RD	82	int desc_size;
	83	int buf_size;
	84	int err;
	85	int i;
	86
	87	/* Sanity check SRQ size before proceeding */
	88	if (init_attr->attr.max_wr >= dev->dev->caps.max_srq_wqes \|\|
	89	init_attr->attr.max_sge > dev->dev->caps.max_srq_sge)
	90	return ERR_PTR(-EINVAL);
	91
	92	srq = kmalloc(sizeof *srq, GFP_KERNEL);
	93	if (!srq)
	94	return ERR_PTR(-ENOMEM);
	95
	96	mutex_init(&srq->mutex);
	97	spin_lock_init(&srq->lock);
	98	srq->msrq.max = roundup_pow_of_two(init_attr->attr.max_wr + 1);
	99	srq->msrq.max_gs = init_attr->attr.max_sge;
	100
	101	desc_size = max(32UL,
	102	roundup_pow_of_two(sizeof (struct mlx4_wqe_srq_next_seg) +
	103	srq->msrq.max_gs *
	104	sizeof (struct mlx4_wqe_data_seg)));
	105	srq->msrq.wqe_shift = ilog2(desc_size);
	106
	107	buf_size = srq->msrq.max * desc_size;
	108
	109	if (pd->uobject) {
	110	struct mlx4_ib_create_srq ucmd;
	111
	112	if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
	113	err = -EFAULT;
	114	goto err_srq;
	115	}
	116
	117	srq->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr,
cb9fbc5c	118	buf_size, 0, 0);
225c7b1f RD	119	if (IS_ERR(srq->umem)) {
	120	err = PTR_ERR(srq->umem);
	121	goto err_srq;
	122	}
	123
	124	err = mlx4_mtt_init(dev->dev, ib_umem_page_count(srq->umem),
	125	ilog2(srq->umem->page_size), &srq->mtt);
	126	if (err)
	127	goto err_buf;
	128
	129	err = mlx4_ib_umem_write_mtt(dev, &srq->mtt, srq->umem);
	130	if (err)
	131	goto err_mtt;
	132
	133	err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context),
	134	ucmd.db_addr, &srq->db);
	135	if (err)
	136	goto err_mtt;
	137	} else {
40f2287b	138	err = mlx4_db_alloc(dev->dev, &srq->db, 0, GFP_KERNEL);
225c7b1f RD	139	if (err)
	140	goto err_srq;
	141
	142	*srq->db.db = 0;
	143
40f2287b JK	144	if (mlx4_buf_alloc(dev->dev, buf_size, PAGE_SIZE * 2, &srq->buf,
40f2287b JK	145	GFP_KERNEL)) {
225c7b1f RD	146	err = -ENOMEM;
	147	goto err_db;
	148	}
	149
	150	srq->head = 0;
	151	srq->tail = srq->msrq.max - 1;
	152	srq->wqe_ctr = 0;
	153
	154	for (i = 0; i < srq->msrq.max; ++i) {
	155	next = get_wqe(srq, i);
	156	next->next_wqe_index =
	157	cpu_to_be16((i + 1) & (srq->msrq.max - 1));
4c425588 JM	158
	159	for (scatter = (void *) (next + 1);
	160	(void ) scatter < (void ) next + desc_size;
	161	++scatter)
	162	scatter->lkey = cpu_to_be32(MLX4_INVALID_LKEY);
225c7b1f RD	163	}
	164
	165	err = mlx4_mtt_init(dev->dev, srq->buf.npages, srq->buf.page_shift,
	166	&srq->mtt);
	167	if (err)
	168	goto err_buf;
	169
40f2287b	170	err = mlx4_buf_write_mtt(dev->dev, &srq->mtt, &srq->buf, GFP_KERNEL);
225c7b1f RD	171	if (err)
	172	goto err_mtt;
	173
	174	srq->wrid = kmalloc(srq->msrq.max * sizeof (u64), GFP_KERNEL);
	175	if (!srq->wrid) {
0ef2f05c WW	176	srq->wrid = __vmalloc(srq->msrq.max * sizeof(u64),
	177	GFP_KERNEL, PAGE_KERNEL);
	178	if (!srq->wrid) {
	179	err = -ENOMEM;
	180	goto err_mtt;
	181	}
225c7b1f RD	182	}
	183	}
	184
18abd5ea SH	185	cqn = (init_attr->srq_type == IB_SRQT_XRC) ?
	186	to_mcq(init_attr->ext.xrc.cq)->mcq.cqn : 0;
	187	xrcdn = (init_attr->srq_type == IB_SRQT_XRC) ?
	188	to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn :
	189	(u16) dev->dev->caps.reserved_xrcds;
	190	err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, cqn, xrcdn, &srq->mtt,
225c7b1f RD	191	srq->db.dma, &srq->msrq);
	192	if (err)
	193	goto err_wrid;
	194
	195	srq->msrq.event = mlx4_ib_srq_event;
18abd5ea	196	srq->ibsrq.ext.xrc.srq_num = srq->msrq.srqn;
225c7b1f RD	197
	198	if (pd->uobject)
	199	if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof (__u32))) {
	200	err = -EFAULT;
	201	goto err_wrid;
	202	}
	203
	204	init_attr->attr.max_wr = srq->msrq.max - 1;
	205
	206	return &srq->ibsrq;
	207
	208	err_wrid:
	209	if (pd->uobject)
	210	mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context), &srq->db);
	211	else
0ef2f05c	212	kvfree(srq->wrid);
225c7b1f RD	213
	214	err_mtt:
	215	mlx4_mtt_cleanup(dev->dev, &srq->mtt);
	216
	217	err_buf:
	218	if (pd->uobject)
	219	ib_umem_release(srq->umem);
	220	else
	221	mlx4_buf_free(dev->dev, buf_size, &srq->buf);
	222
	223	err_db:
	224	if (!pd->uobject)
6296883c	225	mlx4_db_free(dev->dev, &srq->db);
225c7b1f RD	226
	227	err_srq:
	228	kfree(srq);
	229
	230	return ERR_PTR(err);
	231	}
	232
	233	int mlx4_ib_modify_srq(struct ib_srq ibsrq, struct ib_srq_attr attr,
	234	enum ib_srq_attr_mask attr_mask, struct ib_udata *udata)
	235	{
	236	struct mlx4_ib_dev *dev = to_mdev(ibsrq->device);
	237	struct mlx4_ib_srq *srq = to_msrq(ibsrq);
	238	int ret;
	239
	240	/* We don't support resizing SRQs (yet?) */
	241	if (attr_mask & IB_SRQ_MAX_WR)
	242	return -EINVAL;
	243
	244	if (attr_mask & IB_SRQ_LIMIT) {
	245	if (attr->srq_limit >= srq->msrq.max)
	246	return -EINVAL;
	247
	248	mutex_lock(&srq->mutex);
	249	ret = mlx4_srq_arm(dev->dev, &srq->msrq, attr->srq_limit);
	250	mutex_unlock(&srq->mutex);
	251
	252	if (ret)
	253	return ret;
	254	}
	255
	256	return 0;
	257	}
	258
65541cb7 JM	259	int mlx4_ib_query_srq(struct ib_srq ibsrq, struct ib_srq_attr srq_attr)
	260	{
	261	struct mlx4_ib_dev *dev = to_mdev(ibsrq->device);
	262	struct mlx4_ib_srq *srq = to_msrq(ibsrq);
	263	int ret;
	264	int limit_watermark;
	265
	266	ret = mlx4_srq_query(dev->dev, &srq->msrq, &limit_watermark);
	267	if (ret)
	268	return ret;
	269
d7dc3ccb	270	srq_attr->srq_limit = limit_watermark;
65541cb7 JM	271	srq_attr->max_wr = srq->msrq.max - 1;
	272	srq_attr->max_sge = srq->msrq.max_gs;
	273
	274	return 0;
	275	}
	276
225c7b1f RD	277	int mlx4_ib_destroy_srq(struct ib_srq *srq)
	278	{
	279	struct mlx4_ib_dev *dev = to_mdev(srq->device);
	280	struct mlx4_ib_srq *msrq = to_msrq(srq);
	281
	282	mlx4_srq_free(dev->dev, &msrq->msrq);
	283	mlx4_mtt_cleanup(dev->dev, &msrq->mtt);
	284
	285	if (srq->uobject) {
	286	mlx4_ib_db_unmap_user(to_mucontext(srq->uobject->context), &msrq->db);
	287	ib_umem_release(msrq->umem);
	288	} else {
	289	kfree(msrq->wrid);
	290	mlx4_buf_free(dev->dev, msrq->msrq.max << msrq->msrq.wqe_shift,
	291	&msrq->buf);
6296883c	292	mlx4_db_free(dev->dev, &msrq->db);
225c7b1f RD	293	}
	294
	295	kfree(msrq);
	296
	297	return 0;
	298	}
	299
	300	void mlx4_ib_free_srq_wqe(struct mlx4_ib_srq *srq, int wqe_index)
	301	{
	302	struct mlx4_wqe_srq_next_seg *next;
	303
	304	/* always called with interrupts disabled. */
	305	spin_lock(&srq->lock);
	306
	307	next = get_wqe(srq, srq->tail);
	308	next->next_wqe_index = cpu_to_be16(wqe_index);
	309	srq->tail = wqe_index;
	310
	311	spin_unlock(&srq->lock);
	312	}
	313
	314	int mlx4_ib_post_srq_recv(struct ib_srq ibsrq, struct ib_recv_wr wr,
	315	struct ib_recv_wr **bad_wr)
	316	{
	317	struct mlx4_ib_srq *srq = to_msrq(ibsrq);
	318	struct mlx4_wqe_srq_next_seg *next;
	319	struct mlx4_wqe_data_seg *scat;
	320	unsigned long flags;
	321	int err = 0;
	322	int nreq;
	323	int i;
35f05dab	324	struct mlx4_ib_dev *mdev = to_mdev(ibsrq->device);
225c7b1f RD	325
225c7b1f RD	326	spin_lock_irqsave(&srq->lock, flags);
35f05dab YH	327	if (mdev->dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
	328	err = -EIO;
	329	*bad_wr = wr;
	330	nreq = 0;
	331	goto out;
	332	}
225c7b1f RD	333
	334	for (nreq = 0; wr; ++nreq, wr = wr->next) {
	335	if (unlikely(wr->num_sge > srq->msrq.max_gs)) {
	336	err = -EINVAL;
	337	*bad_wr = wr;
	338	break;
	339	}
	340
56a8c8b6 RD	341	if (unlikely(srq->head == srq->tail)) {
	342	err = -ENOMEM;
	343	*bad_wr = wr;
	344	break;
	345	}
	346
225c7b1f RD	347	srq->wrid[srq->head] = wr->wr_id;
	348
	349	next = get_wqe(srq, srq->head);
	350	srq->head = be16_to_cpu(next->next_wqe_index);
	351	scat = (struct mlx4_wqe_data_seg *) (next + 1);
	352
	353	for (i = 0; i < wr->num_sge; ++i) {
	354	scat[i].byte_count = cpu_to_be32(wr->sg_list[i].length);
	355	scat[i].lkey = cpu_to_be32(wr->sg_list[i].lkey);
	356	scat[i].addr = cpu_to_be64(wr->sg_list[i].addr);
	357	}
	358
	359	if (i < srq->msrq.max_gs) {
	360	scat[i].byte_count = 0;
	361	scat[i].lkey = cpu_to_be32(MLX4_INVALID_LKEY);
	362	scat[i].addr = 0;
	363	}
	364	}
	365
	366	if (likely(nreq)) {
	367	srq->wqe_ctr += nreq;
	368
	369	/*
	370	* Make sure that descriptors are written before
	371	* doorbell record.
	372	*/
	373	wmb();
	374
	375	*srq->db.db = cpu_to_be32(srq->wqe_ctr);
	376	}
35f05dab	377	out:
225c7b1f RD	378
	379	spin_unlock_irqrestore(&srq->lock, flags);
	380
	381	return err;
	382	}