[linux-2.6-block.git] / drivers / infiniband / hw / mlx5 / srq.c

/*
 * Copyright (c) 2013-2015, 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/module.h>
#include <linux/mlx5/qp.h>
#include <linux/mlx5/srq.h>
#include <linux/slab.h>
#include <rdma/ib_umem.h>
#include <rdma/ib_user_verbs.h>

#include "mlx5_ib.h"

/* not supported currently */
static int srq_signature;

static void *get_wqe(struct mlx5_ib_srq *srq, int n)
{
	return mlx5_buf_offset(&srq->buf, n << srq->msrq.wqe_shift);
}

static void mlx5_ib_srq_event(struct mlx5_core_srq *srq, enum mlx5_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 MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
			break;
		case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
			event.event = IB_EVENT_SRQ_ERR;
			break;
		default:
			pr_warn("mlx5_ib: Unexpected event type %d on SRQ %06x\n",
				type, srq->srqn);
			return;
		}

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

static int create_srq_user(struct ib_pd *pd, struct mlx5_ib_srq *srq,
			   struct mlx5_srq_attr *in,
			   struct ib_udata *udata, int buf_size)
{
	struct mlx5_ib_dev *dev = to_mdev(pd->device);
	struct mlx5_ib_create_srq ucmd = {};
	size_t ucmdlen;
	int err;
	int npages;
	int page_shift;
	int ncont;
	u32 offset;
	u32 uidx = MLX5_IB_DEFAULT_UIDX;

	ucmdlen = min(udata->inlen, sizeof(ucmd));

	if (ib_copy_from_udata(&ucmd, udata, ucmdlen)) {
		mlx5_ib_dbg(dev, "failed copy udata\n");
		return -EFAULT;
	}

	if (ucmd.reserved0 || ucmd.reserved1)
		return -EINVAL;

	if (udata->inlen > sizeof(ucmd) &&
	    !ib_is_udata_cleared(udata, sizeof(ucmd),
				 udata->inlen - sizeof(ucmd)))
		return -EINVAL;

	if (in->type != IB_SRQT_BASIC) {
		err = get_srq_user_index(to_mucontext(pd->uobject->context),
					 &ucmd, udata->inlen, &uidx);
		if (err)
			return err;
	}

	srq->wq_sig = !!(ucmd.flags & MLX5_SRQ_FLAG_SIGNATURE);

	srq->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr, buf_size,
				0, 0);
	if (IS_ERR(srq->umem)) {
		mlx5_ib_dbg(dev, "failed umem get, size %d\n", buf_size);
		err = PTR_ERR(srq->umem);
		return err;
	}

	mlx5_ib_cont_pages(srq->umem, ucmd.buf_addr, 0, &npages,
			   &page_shift, &ncont, NULL);
	err = mlx5_ib_get_buf_offset(ucmd.buf_addr, page_shift,
				     &offset);
	if (err) {
		mlx5_ib_warn(dev, "bad offset\n");
		goto err_umem;
	}

	in->pas = kvzalloc(sizeof(*in->pas) * ncont, GFP_KERNEL);
	if (!in->pas) {
		err = -ENOMEM;
		goto err_umem;
	}

	mlx5_ib_populate_pas(dev, srq->umem, page_shift, in->pas, 0);

	err = mlx5_ib_db_map_user(to_mucontext(pd->uobject->context),
				  ucmd.db_addr, &srq->db);
	if (err) {
		mlx5_ib_dbg(dev, "map doorbell failed\n");
		goto err_in;
	}

	in->log_page_size = page_shift - MLX5_ADAPTER_PAGE_SHIFT;
	in->page_offset = offset;
	if (MLX5_CAP_GEN(dev->mdev, cqe_version) == MLX5_CQE_VERSION_V1 &&
	    in->type != IB_SRQT_BASIC)
		in->user_index = uidx;

	return 0;

err_in:
	kvfree(in->pas);

err_umem:
	ib_umem_release(srq->umem);

	return err;
}

static int create_srq_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_srq *srq,
			     struct mlx5_srq_attr *in, int buf_size)
{
	int err;
	int i;
	struct mlx5_wqe_srq_next_seg *next;

	err = mlx5_db_alloc(dev->mdev, &srq->db);
	if (err) {
		mlx5_ib_warn(dev, "alloc dbell rec failed\n");
		return err;
	}

	if (mlx5_buf_alloc(dev->mdev, buf_size, &srq->buf)) {
		mlx5_ib_dbg(dev, "buf alloc failed\n");
		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));
	}

	mlx5_ib_dbg(dev, "srq->buf.page_shift = %d\n", srq->buf.page_shift);
	in->pas = kvzalloc(sizeof(*in->pas) * srq->buf.npages, GFP_KERNEL);
	if (!in->pas) {
		err = -ENOMEM;
		goto err_buf;
	}
	mlx5_fill_page_array(&srq->buf, in->pas);

	srq->wrid = kvmalloc_array(srq->msrq.max, sizeof(u64), GFP_KERNEL);
	if (!srq->wrid) {
		err = -ENOMEM;
		goto err_in;
	}
	srq->wq_sig = !!srq_signature;

	in->log_page_size = srq->buf.page_shift - MLX5_ADAPTER_PAGE_SHIFT;
	if (MLX5_CAP_GEN(dev->mdev, cqe_version) == MLX5_CQE_VERSION_V1 &&
	    in->type != IB_SRQT_BASIC)
		in->user_index = MLX5_IB_DEFAULT_UIDX;

	return 0;

err_in:
	kvfree(in->pas);

err_buf:
	mlx5_buf_free(dev->mdev, &srq->buf);

err_db:
	mlx5_db_free(dev->mdev, &srq->db);
	return err;
}

static void destroy_srq_user(struct ib_pd *pd, struct mlx5_ib_srq *srq)
{
	mlx5_ib_db_unmap_user(to_mucontext(pd->uobject->context), &srq->db);
	ib_umem_release(srq->umem);
}


static void destroy_srq_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_srq *srq)
{
	kvfree(srq->wrid);
	mlx5_buf_free(dev->mdev, &srq->buf);
	mlx5_db_free(dev->mdev, &srq->db);
}

struct ib_srq *mlx5_ib_create_srq(struct ib_pd *pd,
				  struct ib_srq_init_attr *init_attr,
				  struct ib_udata *udata)
{
	struct mlx5_ib_dev *dev = to_mdev(pd->device);
	struct mlx5_ib_srq *srq;
	size_t desc_size;
	size_t buf_size;
	int err;
	struct mlx5_srq_attr in = {0};
	__u32 max_srq_wqes = 1 << MLX5_CAP_GEN(dev->mdev, log_max_srq_sz);

	/* Sanity check SRQ size before proceeding */
	if (init_attr->attr.max_wr >= max_srq_wqes) {
		mlx5_ib_dbg(dev, "max_wr %d, cap %d\n",
			    init_attr->attr.max_wr,
			    max_srq_wqes);
		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 = sizeof(struct mlx5_wqe_srq_next_seg) +
		    srq->msrq.max_gs * sizeof(struct mlx5_wqe_data_seg);
	if (desc_size == 0 || srq->msrq.max_gs > desc_size)
		return ERR_PTR(-EINVAL);
	desc_size = roundup_pow_of_two(desc_size);
	desc_size = max_t(size_t, 32, desc_size);
	if (desc_size < sizeof(struct mlx5_wqe_srq_next_seg))
		return ERR_PTR(-EINVAL);
	srq->msrq.max_avail_gather = (desc_size - sizeof(struct mlx5_wqe_srq_next_seg)) /
		sizeof(struct mlx5_wqe_data_seg);
	srq->msrq.wqe_shift = ilog2(desc_size);
	buf_size = srq->msrq.max * desc_size;
	if (buf_size < desc_size)
		return ERR_PTR(-EINVAL);
	in.type = init_attr->srq_type;

	if (pd->uobject)
		err = create_srq_user(pd, srq, &in, udata, buf_size);
	else
		err = create_srq_kernel(dev, srq, &in, buf_size);

	if (err) {
		mlx5_ib_warn(dev, "create srq %s failed, err %d\n",
			     pd->uobject ? "user" : "kernel", err);
		goto err_srq;
	}

	in.log_size = ilog2(srq->msrq.max);
	in.wqe_shift = srq->msrq.wqe_shift - 4;
	if (srq->wq_sig)
		in.flags |= MLX5_SRQ_FLAG_WQ_SIG;

	if (init_attr->srq_type == IB_SRQT_XRC)
		in.xrcd = to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn;
	else
		in.xrcd = to_mxrcd(dev->devr.x0)->xrcdn;

	if (init_attr->srq_type == IB_SRQT_TM) {
		in.tm_log_list_size =
			ilog2(init_attr->ext.tag_matching.max_num_tags) + 1;
		if (in.tm_log_list_size >
		    MLX5_CAP_GEN(dev->mdev, log_tag_matching_list_sz)) {
			mlx5_ib_dbg(dev, "TM SRQ max_num_tags exceeding limit\n");
			err = -EINVAL;
			goto err_usr_kern_srq;
		}
		in.flags |= MLX5_SRQ_FLAG_RNDV;
	}

	if (ib_srq_has_cq(init_attr->srq_type))
		in.cqn = to_mcq(init_attr->ext.cq)->mcq.cqn;
	else
		in.cqn = to_mcq(dev->devr.c0)->mcq.cqn;

	in.pd = to_mpd(pd)->pdn;
	in.db_record = srq->db.dma;
	err = mlx5_core_create_srq(dev->mdev, &srq->msrq, &in);
	kvfree(in.pas);
	if (err) {
		mlx5_ib_dbg(dev, "create SRQ failed, err %d\n", err);
		goto err_usr_kern_srq;
	}

	mlx5_ib_dbg(dev, "create SRQ with srqn 0x%x\n", srq->msrq.srqn);

	srq->msrq.event = mlx5_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))) {
			mlx5_ib_dbg(dev, "copy to user failed\n");
			err = -EFAULT;
			goto err_core;
		}

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

	return &srq->ibsrq;

err_core:
	mlx5_core_destroy_srq(dev->mdev, &srq->msrq);

err_usr_kern_srq:
	if (pd->uobject)
		destroy_srq_user(pd, srq);
	else
		destroy_srq_kernel(dev, srq);

err_srq:
	kfree(srq);

	return ERR_PTR(err);
}

int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata)
{
	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
	struct mlx5_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 = mlx5_core_arm_srq(dev->mdev, &srq->msrq, attr->srq_limit, 1);
		mutex_unlock(&srq->mutex);

		if (ret)
			return ret;
	}

	return 0;
}

int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
{
	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
	int ret;
	struct mlx5_srq_attr *out;

	out = kzalloc(sizeof(*out), GFP_KERNEL);
	if (!out)
		return -ENOMEM;

	ret = mlx5_core_query_srq(dev->mdev, &srq->msrq, out);
	if (ret)
		goto out_box;

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

out_box:
	kfree(out);
	return ret;
}

int mlx5_ib_destroy_srq(struct ib_srq *srq)
{
	struct mlx5_ib_dev *dev = to_mdev(srq->device);
	struct mlx5_ib_srq *msrq = to_msrq(srq);

	mlx5_core_destroy_srq(dev->mdev, &msrq->msrq);

	if (srq->uobject) {
		mlx5_ib_db_unmap_user(to_mucontext(srq->uobject->context), &msrq->db);
		ib_umem_release(msrq->umem);
	} else {
		destroy_srq_kernel(dev, msrq);
	}

	kfree(srq);
	return 0;
}

void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index)
{
	struct mlx5_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 mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
			  struct ib_recv_wr **bad_wr)
{
	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
	struct mlx5_wqe_srq_next_seg *next;
	struct mlx5_wqe_data_seg *scat;
	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
	struct mlx5_core_dev *mdev = dev->mdev;
	unsigned long flags;
	int err = 0;
	int nreq;
	int i;

	spin_lock_irqsave(&srq->lock, flags);

	if (mdev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
		err = -EIO;
		*bad_wr = wr;
		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 mlx5_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_avail_gather) {
			scat[i].byte_count = 0;
			scat[i].lkey       = cpu_to_be32(MLX5_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
e126ba97	1	/*
6cf0a15f	2	* Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
e126ba97 EC	3	*
	4	* This software is available to you under a choice of one of two
	5	* licenses. You may choose to be licensed under the terms of the GNU
	6	* General Public License (GPL) Version 2, available from the file
	7	* COPYING in the main directory of this source tree, or the
	8	* OpenIB.org BSD license below:
	9	*
	10	* Redistribution and use in source and binary forms, with or
	11	* without modification, are permitted provided that the following
	12	* conditions are met:
	13	*
	14	* - Redistributions of source code must retain the above
	15	* copyright notice, this list of conditions and the following
	16	* disclaimer.
	17	*
	18	* - Redistributions in binary form must reproduce the above
	19	* copyright notice, this list of conditions and the following
	20	* disclaimer in the documentation and/or other materials
	21	* provided with the distribution.
	22	*
	23	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
	24	* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
	25	* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
	26	* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
	27	* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
	28	* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
	29	* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
	30	* SOFTWARE.
	31	*/
	32
	33	#include <linux/module.h>
	34	#include <linux/mlx5/qp.h>
	35	#include <linux/mlx5/srq.h>
	36	#include <linux/slab.h>
	37	#include <rdma/ib_umem.h>
43bc8893	38	#include <rdma/ib_user_verbs.h>
e126ba97 EC	39
e126ba97 EC	40	#include "mlx5_ib.h"
e126ba97 EC	41
	42	/* not supported currently */
	43	static int srq_signature;
	44
	45	static void get_wqe(struct mlx5_ib_srq srq, int n)
	46	{
	47	return mlx5_buf_offset(&srq->buf, n << srq->msrq.wqe_shift);
	48	}
	49
	50	static void mlx5_ib_srq_event(struct mlx5_core_srq *srq, enum mlx5_event type)
	51	{
	52	struct ib_event event;
	53	struct ib_srq *ibsrq = &to_mibsrq(srq)->ibsrq;
	54
	55	if (ibsrq->event_handler) {
	56	event.device = ibsrq->device;
	57	event.element.srq = ibsrq;
	58	switch (type) {
	59	case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
	60	event.event = IB_EVENT_SRQ_LIMIT_REACHED;
	61	break;
	62	case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
	63	event.event = IB_EVENT_SRQ_ERR;
	64	break;
	65	default:
	66	pr_warn("mlx5_ib: Unexpected event type %d on SRQ %06x\n",
	67	type, srq->srqn);
	68	return;
	69	}
	70
	71	ibsrq->event_handler(&event, ibsrq->srq_context);
	72	}
	73	}
	74
	75	static int create_srq_user(struct ib_pd pd, struct mlx5_ib_srq srq,
af1ba291 AK	76	struct mlx5_srq_attr *in,
af1ba291 AK	77	struct ib_udata *udata, int buf_size)
e126ba97 EC	78	{
e126ba97 EC	79	struct mlx5_ib_dev *dev = to_mdev(pd->device);
cfb5e088	80	struct mlx5_ib_create_srq ucmd = {};
43bc8893	81	size_t ucmdlen;
e126ba97 EC	82	int err;
	83	int npages;
	84	int page_shift;
	85	int ncont;
	86	u32 offset;
cfb5e088	87	u32 uidx = MLX5_IB_DEFAULT_UIDX;
e126ba97	88
3d943c9d	89	ucmdlen = min(udata->inlen, sizeof(ucmd));
43bc8893 YD	90
43bc8893 YD	91	if (ib_copy_from_udata(&ucmd, udata, ucmdlen)) {
e126ba97 EC	92	mlx5_ib_dbg(dev, "failed copy udata\n");
	93	return -EFAULT;
	94	}
43bc8893	95
cfb5e088	96	if (ucmd.reserved0 \|\| ucmd.reserved1)
43bc8893 YD	97	return -EINVAL;
43bc8893 YD	98
3d943c9d	99	if (udata->inlen > sizeof(ucmd) &&
cfb5e088	100	!ib_is_udata_cleared(udata, sizeof(ucmd),
3d943c9d	101	udata->inlen - sizeof(ucmd)))
cfb5e088 HA	102	return -EINVAL;
cfb5e088 HA	103
3fd3307e	104	if (in->type != IB_SRQT_BASIC) {
85d9691c MD	105	err = get_srq_user_index(to_mucontext(pd->uobject->context),
	106	&ucmd, udata->inlen, &uidx);
	107	if (err)
	108	return err;
	109	}
cfb5e088	110
e126ba97 EC	111	srq->wq_sig = !!(ucmd.flags & MLX5_SRQ_FLAG_SIGNATURE);
	112
	113	srq->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr, buf_size,
	114	0, 0);
	115	if (IS_ERR(srq->umem)) {
	116	mlx5_ib_dbg(dev, "failed umem get, size %d\n", buf_size);
	117	err = PTR_ERR(srq->umem);
	118	return err;
	119	}
	120
762f899a	121	mlx5_ib_cont_pages(srq->umem, ucmd.buf_addr, 0, &npages,
e126ba97 EC	122	&page_shift, &ncont, NULL);
	123	err = mlx5_ib_get_buf_offset(ucmd.buf_addr, page_shift,
	124	&offset);
	125	if (err) {
	126	mlx5_ib_warn(dev, "bad offset\n");
	127	goto err_umem;
	128	}
	129
1b9a07ee	130	in->pas = kvzalloc(sizeof(in->pas) ncont, GFP_KERNEL);
af1ba291	131	if (!in->pas) {
e126ba97 EC	132	err = -ENOMEM;
	133	goto err_umem;
	134	}
	135
af1ba291	136	mlx5_ib_populate_pas(dev, srq->umem, page_shift, in->pas, 0);
e126ba97 EC	137
	138	err = mlx5_ib_db_map_user(to_mucontext(pd->uobject->context),
	139	ucmd.db_addr, &srq->db);
	140	if (err) {
	141	mlx5_ib_dbg(dev, "map doorbell failed\n");
	142	goto err_in;
	143	}
	144
af1ba291 AK	145	in->log_page_size = page_shift - MLX5_ADAPTER_PAGE_SHIFT;
	146	in->page_offset = offset;
	147	if (MLX5_CAP_GEN(dev->mdev, cqe_version) == MLX5_CQE_VERSION_V1 &&
3fd3307e	148	in->type != IB_SRQT_BASIC)
af1ba291	149	in->user_index = uidx;
cfb5e088	150
e126ba97 EC	151	return 0;
	152
	153	err_in:
af1ba291	154	kvfree(in->pas);
e126ba97 EC	155
	156	err_umem:
	157	ib_umem_release(srq->umem);
	158
	159	return err;
	160	}
	161
	162	static int create_srq_kernel(struct mlx5_ib_dev dev, struct mlx5_ib_srq srq,
af1ba291	163	struct mlx5_srq_attr *in, int buf_size)
e126ba97 EC	164	{
	165	int err;
	166	int i;
	167	struct mlx5_wqe_srq_next_seg *next;
e126ba97	168
9603b61d	169	err = mlx5_db_alloc(dev->mdev, &srq->db);
e126ba97 EC	170	if (err) {
	171	mlx5_ib_warn(dev, "alloc dbell rec failed\n");
	172	return err;
	173	}
	174
64ffaa21	175	if (mlx5_buf_alloc(dev->mdev, buf_size, &srq->buf)) {
e126ba97 EC	176	mlx5_ib_dbg(dev, "buf alloc failed\n");
	177	err = -ENOMEM;
	178	goto err_db;
	179	}
e126ba97 EC	180
	181	srq->head = 0;
	182	srq->tail = srq->msrq.max - 1;
	183	srq->wqe_ctr = 0;
	184
	185	for (i = 0; i < srq->msrq.max; i++) {
	186	next = get_wqe(srq, i);
	187	next->next_wqe_index =
	188	cpu_to_be16((i + 1) & (srq->msrq.max - 1));
	189	}
	190
0fd27a88	191	mlx5_ib_dbg(dev, "srq->buf.page_shift = %d\n", srq->buf.page_shift);
1b9a07ee	192	in->pas = kvzalloc(sizeof(in->pas) srq->buf.npages, GFP_KERNEL);
af1ba291	193	if (!in->pas) {
e126ba97 EC	194	err = -ENOMEM;
	195	goto err_buf;
	196	}
af1ba291	197	mlx5_fill_page_array(&srq->buf, in->pas);
e126ba97	198
b5883008	199	srq->wrid = kvmalloc_array(srq->msrq.max, sizeof(u64), GFP_KERNEL);
e126ba97	200	if (!srq->wrid) {
e126ba97 EC	201	err = -ENOMEM;
	202	goto err_in;
	203	}
	204	srq->wq_sig = !!srq_signature;
	205
0fd27a88	206	in->log_page_size = srq->buf.page_shift - MLX5_ADAPTER_PAGE_SHIFT;
af1ba291	207	if (MLX5_CAP_GEN(dev->mdev, cqe_version) == MLX5_CQE_VERSION_V1 &&
3fd3307e	208	in->type != IB_SRQT_BASIC)
af1ba291	209	in->user_index = MLX5_IB_DEFAULT_UIDX;
cfb5e088	210
e126ba97 EC	211	return 0;
	212
	213	err_in:
af1ba291	214	kvfree(in->pas);
e126ba97 EC	215
e126ba97 EC	216	err_buf:
9603b61d	217	mlx5_buf_free(dev->mdev, &srq->buf);
e126ba97 EC	218
e126ba97 EC	219	err_db:
9603b61d	220	mlx5_db_free(dev->mdev, &srq->db);
e126ba97 EC	221	return err;
	222	}
	223
	224	static void destroy_srq_user(struct ib_pd pd, struct mlx5_ib_srq srq)
	225	{
	226	mlx5_ib_db_unmap_user(to_mucontext(pd->uobject->context), &srq->db);
	227	ib_umem_release(srq->umem);
	228	}
	229
	230
	231	static void destroy_srq_kernel(struct mlx5_ib_dev dev, struct mlx5_ib_srq srq)
	232	{
b5883008	233	kvfree(srq->wrid);
9603b61d JM	234	mlx5_buf_free(dev->mdev, &srq->buf);
9603b61d JM	235	mlx5_db_free(dev->mdev, &srq->db);
e126ba97 EC	236	}
	237
	238	struct ib_srq mlx5_ib_create_srq(struct ib_pd pd,
	239	struct ib_srq_init_attr *init_attr,
	240	struct ib_udata *udata)
	241	{
	242	struct mlx5_ib_dev *dev = to_mdev(pd->device);
	243	struct mlx5_ib_srq *srq;
c2b37f76 BP	244	size_t desc_size;
c2b37f76 BP	245	size_t buf_size;
e126ba97	246	int err;
af1ba291	247	struct mlx5_srq_attr in = {0};
938fe83c	248	__u32 max_srq_wqes = 1 << MLX5_CAP_GEN(dev->mdev, log_max_srq_sz);
e126ba97 EC	249
e126ba97 EC	250	/* Sanity check SRQ size before proceeding */
938fe83c	251	if (init_attr->attr.max_wr >= max_srq_wqes) {
e126ba97 EC	252	mlx5_ib_dbg(dev, "max_wr %d, cap %d\n",
e126ba97 EC	253	init_attr->attr.max_wr,
938fe83c	254	max_srq_wqes);
e126ba97 EC	255	return ERR_PTR(-EINVAL);
	256	}
	257
	258	srq = kmalloc(sizeof(*srq), GFP_KERNEL);
	259	if (!srq)
	260	return ERR_PTR(-ENOMEM);
	261
	262	mutex_init(&srq->mutex);
	263	spin_lock_init(&srq->lock);
	264	srq->msrq.max = roundup_pow_of_two(init_attr->attr.max_wr + 1);
	265	srq->msrq.max_gs = init_attr->attr.max_sge;
	266
	267	desc_size = sizeof(struct mlx5_wqe_srq_next_seg) +
	268	srq->msrq.max_gs * sizeof(struct mlx5_wqe_data_seg);
c2b37f76 BP	269	if (desc_size == 0 \|\| srq->msrq.max_gs > desc_size)
c2b37f76 BP	270	return ERR_PTR(-EINVAL);
e126ba97	271	desc_size = roundup_pow_of_two(desc_size);
c2b37f76 BP	272	desc_size = max_t(size_t, 32, desc_size);
	273	if (desc_size < sizeof(struct mlx5_wqe_srq_next_seg))
	274	return ERR_PTR(-EINVAL);
e126ba97 EC	275	srq->msrq.max_avail_gather = (desc_size - sizeof(struct mlx5_wqe_srq_next_seg)) /
	276	sizeof(struct mlx5_wqe_data_seg);
	277	srq->msrq.wqe_shift = ilog2(desc_size);
	278	buf_size = srq->msrq.max * desc_size;
c2b37f76 BP	279	if (buf_size < desc_size)
c2b37f76 BP	280	return ERR_PTR(-EINVAL);
c73b7911	281	in.type = init_attr->srq_type;
e126ba97 EC	282
e126ba97 EC	283	if (pd->uobject)
af1ba291	284	err = create_srq_user(pd, srq, &in, udata, buf_size);
e126ba97	285	else
af1ba291	286	err = create_srq_kernel(dev, srq, &in, buf_size);
e126ba97 EC	287
	288	if (err) {
	289	mlx5_ib_warn(dev, "create srq %s failed, err %d\n",
	290	pd->uobject ? "user" : "kernel", err);
	291	goto err_srq;
	292	}
	293
af1ba291 AK	294	in.log_size = ilog2(srq->msrq.max);
	295	in.wqe_shift = srq->msrq.wqe_shift - 4;
	296	if (srq->wq_sig)
	297	in.flags \|= MLX5_SRQ_FLAG_WQ_SIG;
1a56ff6d AK	298
1a56ff6d AK	299	if (init_attr->srq_type == IB_SRQT_XRC)
af1ba291	300	in.xrcd = to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn;
1a56ff6d	301	else
af1ba291	302	in.xrcd = to_mxrcd(dev->devr.x0)->xrcdn;
1a56ff6d	303
3fd3307e AK	304	if (init_attr->srq_type == IB_SRQT_TM) {
	305	in.tm_log_list_size =
	306	ilog2(init_attr->ext.tag_matching.max_num_tags) + 1;
	307	if (in.tm_log_list_size >
	308	MLX5_CAP_GEN(dev->mdev, log_tag_matching_list_sz)) {
	309	mlx5_ib_dbg(dev, "TM SRQ max_num_tags exceeding limit\n");
	310	err = -EINVAL;
	311	goto err_usr_kern_srq;
	312	}
	313	in.flags \|= MLX5_SRQ_FLAG_RNDV;
	314	}
	315
1a56ff6d AK	316	if (ib_srq_has_cq(init_attr->srq_type))
	317	in.cqn = to_mcq(init_attr->ext.cq)->mcq.cqn;
	318	else
af1ba291	319	in.cqn = to_mcq(dev->devr.c0)->mcq.cqn;
e126ba97	320
af1ba291 AK	321	in.pd = to_mpd(pd)->pdn;
	322	in.db_record = srq->db.dma;
	323	err = mlx5_core_create_srq(dev->mdev, &srq->msrq, &in);
	324	kvfree(in.pas);
e126ba97 EC	325	if (err) {
e126ba97 EC	326	mlx5_ib_dbg(dev, "create SRQ failed, err %d\n", err);
56e1ab0f	327	goto err_usr_kern_srq;
e126ba97 EC	328	}
	329
	330	mlx5_ib_dbg(dev, "create SRQ with srqn 0x%x\n", srq->msrq.srqn);
	331
	332	srq->msrq.event = mlx5_ib_srq_event;
	333	srq->ibsrq.ext.xrc.srq_num = srq->msrq.srqn;
	334
	335	if (pd->uobject)
	336	if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof(__u32))) {
	337	mlx5_ib_dbg(dev, "copy to user failed\n");
	338	err = -EFAULT;
	339	goto err_core;
	340	}
	341
	342	init_attr->attr.max_wr = srq->msrq.max - 1;
	343
	344	return &srq->ibsrq;
	345
	346	err_core:
9603b61d	347	mlx5_core_destroy_srq(dev->mdev, &srq->msrq);
56e1ab0f ML	348
56e1ab0f ML	349	err_usr_kern_srq:
e126ba97 EC	350	if (pd->uobject)
	351	destroy_srq_user(pd, srq);
	352	else
	353	destroy_srq_kernel(dev, srq);
	354
	355	err_srq:
	356	kfree(srq);
	357
	358	return ERR_PTR(err);
	359	}
	360
	361	int mlx5_ib_modify_srq(struct ib_srq ibsrq, struct ib_srq_attr attr,
	362	enum ib_srq_attr_mask attr_mask, struct ib_udata *udata)
	363	{
	364	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
	365	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
	366	int ret;
	367
	368	/* We don't support resizing SRQs yet */
	369	if (attr_mask & IB_SRQ_MAX_WR)
	370	return -EINVAL;
	371
	372	if (attr_mask & IB_SRQ_LIMIT) {
	373	if (attr->srq_limit >= srq->msrq.max)
	374	return -EINVAL;
	375
	376	mutex_lock(&srq->mutex);
9603b61d	377	ret = mlx5_core_arm_srq(dev->mdev, &srq->msrq, attr->srq_limit, 1);
e126ba97 EC	378	mutex_unlock(&srq->mutex);
	379
	380	if (ret)
	381	return ret;
	382	}
	383
	384	return 0;
	385	}
	386
	387	int mlx5_ib_query_srq(struct ib_srq ibsrq, struct ib_srq_attr srq_attr)
	388	{
	389	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
	390	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
	391	int ret;
af1ba291	392	struct mlx5_srq_attr *out;
e126ba97 EC	393
	394	out = kzalloc(sizeof(*out), GFP_KERNEL);
	395	if (!out)
	396	return -ENOMEM;
	397
9603b61d	398	ret = mlx5_core_query_srq(dev->mdev, &srq->msrq, out);
e126ba97 EC	399	if (ret)
	400	goto out_box;
	401
af1ba291	402	srq_attr->srq_limit = out->lwm;
e126ba97 EC	403	srq_attr->max_wr = srq->msrq.max - 1;
	404	srq_attr->max_sge = srq->msrq.max_gs;
	405
	406	out_box:
	407	kfree(out);
	408	return ret;
	409	}
	410
	411	int mlx5_ib_destroy_srq(struct ib_srq *srq)
	412	{
	413	struct mlx5_ib_dev *dev = to_mdev(srq->device);
	414	struct mlx5_ib_srq *msrq = to_msrq(srq);
	415
9603b61d	416	mlx5_core_destroy_srq(dev->mdev, &msrq->msrq);
e126ba97 EC	417
	418	if (srq->uobject) {
	419	mlx5_ib_db_unmap_user(to_mucontext(srq->uobject->context), &msrq->db);
	420	ib_umem_release(msrq->umem);
	421	} else {
1faacf82	422	destroy_srq_kernel(dev, msrq);
e126ba97 EC	423	}
	424
	425	kfree(srq);
	426	return 0;
	427	}
	428
	429	void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index)
	430	{
	431	struct mlx5_wqe_srq_next_seg *next;
	432
	433	/* always called with interrupts disabled. */
	434	spin_lock(&srq->lock);
	435
	436	next = get_wqe(srq, srq->tail);
	437	next->next_wqe_index = cpu_to_be16(wqe_index);
	438	srq->tail = wqe_index;
	439
	440	spin_unlock(&srq->lock);
	441	}
	442
	443	int mlx5_ib_post_srq_recv(struct ib_srq ibsrq, struct ib_recv_wr wr,
	444	struct ib_recv_wr **bad_wr)
	445	{
	446	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
	447	struct mlx5_wqe_srq_next_seg *next;
	448	struct mlx5_wqe_data_seg *scat;
89ea94a7 MG	449	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
89ea94a7 MG	450	struct mlx5_core_dev *mdev = dev->mdev;
e126ba97 EC	451	unsigned long flags;
	452	int err = 0;
	453	int nreq;
	454	int i;
	455
	456	spin_lock_irqsave(&srq->lock, flags);
	457
89ea94a7 MG	458	if (mdev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
	459	err = -EIO;
	460	*bad_wr = wr;
	461	goto out;
	462	}
	463
e126ba97 EC	464	for (nreq = 0; wr; nreq++, wr = wr->next) {
	465	if (unlikely(wr->num_sge > srq->msrq.max_gs)) {
	466	err = -EINVAL;
	467	*bad_wr = wr;
	468	break;
	469	}
	470
	471	if (unlikely(srq->head == srq->tail)) {
	472	err = -ENOMEM;
	473	*bad_wr = wr;
	474	break;
	475	}
	476
	477	srq->wrid[srq->head] = wr->wr_id;
	478
	479	next = get_wqe(srq, srq->head);
	480	srq->head = be16_to_cpu(next->next_wqe_index);
	481	scat = (struct mlx5_wqe_data_seg *)(next + 1);
	482
	483	for (i = 0; i < wr->num_sge; i++) {
	484	scat[i].byte_count = cpu_to_be32(wr->sg_list[i].length);
	485	scat[i].lkey = cpu_to_be32(wr->sg_list[i].lkey);
	486	scat[i].addr = cpu_to_be64(wr->sg_list[i].addr);
	487	}
	488
	489	if (i < srq->msrq.max_avail_gather) {
	490	scat[i].byte_count = 0;
	491	scat[i].lkey = cpu_to_be32(MLX5_INVALID_LKEY);
	492	scat[i].addr = 0;
	493	}
	494	}
	495
	496	if (likely(nreq)) {
	497	srq->wqe_ctr += nreq;
	498
	499	/* Make sure that descriptors are written before
	500	* doorbell record.
	501	*/
	502	wmb();
	503
	504	*srq->db.db = cpu_to_be32(srq->wqe_ctr);
	505	}
89ea94a7	506	out:
e126ba97 EC	507	spin_unlock_irqrestore(&srq->lock, flags);
	508
	509	return err;
	510	}