[mirror_ubuntu-hirsute-kernel.git] / drivers / infiniband / hw / mlx5 / srq.c

// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
/*
 * Copyright (c) 2013-2018, Mellanox Technologies inc.  All rights reserved.
 */

#include <linux/module.h>
#include <linux/mlx5/qp.h>
#include <linux/slab.h>
#include <rdma/ib_umem.h>
#include <rdma/ib_user_verbs.h>
#include "mlx5_ib.h"
#include "srq.h"

static void *get_wqe(struct mlx5_ib_srq *srq, int n)
{
	return mlx5_frag_buf_get_wqe(&srq->fbc, n);
}

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 = {};
	struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context(
		udata, struct mlx5_ib_ucontext, ibucontext);
	size_t ucmdlen;
	int err;
	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(ucontext, &ucmd, udata->inlen, &uidx);
		if (err)
			return err;
	}

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

	srq->umem = ib_umem_get_peer(pd->device, 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;
	}
	in->umem = srq->umem;

	err = mlx5_ib_db_map_user(ucontext, udata, ucmd.db_addr, &srq->db);
	if (err) {
		mlx5_ib_dbg(dev, "map doorbell failed\n");
		goto err_umem;
	}

	in->uid = (in->type != IB_SRQT_XRC) ?  to_mpd(pd)->uid : 0;
	if (MLX5_CAP_GEN(dev->mdev, cqe_version) == MLX5_CQE_VERSION_V1 &&
	    in->type != IB_SRQT_BASIC)
		in->user_index = uidx;

	return 0;

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_frag_buf_alloc_node(dev->mdev, buf_size, &srq->buf,
				     dev->mdev->priv.numa_node)) {
		mlx5_ib_dbg(dev, "buf alloc failed\n");
		err = -ENOMEM;
		goto err_db;
	}

	mlx5_init_fbc(srq->buf.frags, srq->msrq.wqe_shift, ilog2(srq->msrq.max),
		      &srq->fbc);

	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 = kvcalloc(srq->buf.npages, sizeof(*in->pas), GFP_KERNEL);
	if (!in->pas) {
		err = -ENOMEM;
		goto err_buf;
	}
	mlx5_fill_page_frag_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 = 0;

	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_frag_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,
			     struct ib_udata *udata)
{
	mlx5_ib_db_unmap_user(
		rdma_udata_to_drv_context(
			udata,
			struct mlx5_ib_ucontext,
			ibucontext),
		&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_frag_buf_free(dev->mdev, &srq->buf);
	mlx5_db_free(dev->mdev, &srq->db);
}

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

	if (init_attr->srq_type != IB_SRQT_BASIC &&
	    init_attr->srq_type != IB_SRQT_XRC &&
	    init_attr->srq_type != IB_SRQT_TM)
		return -EOPNOTSUPP;

	/* 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 -EINVAL;
	}

	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 -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 -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 -EINVAL;

	in.type = init_attr->srq_type;

	if (udata)
		err = create_srq_user(ib_srq->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",
			     udata ? "user" : "kernel", err);
		return err;
	}

	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 && init_attr->ext.xrc.xrcd)
		in.xrcd = to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn;
	else
		in.xrcd = dev->devr.xrcdn0;

	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(ib_srq->pd)->pdn;
	in.db_record = srq->db.dma;
	err = mlx5_cmd_create_srq(dev, &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 (udata) {
		struct mlx5_ib_create_srq_resp resp = {
			.srqn = srq->msrq.srqn,
		};

		if (ib_copy_to_udata(udata, &resp, min(udata->outlen,
				     sizeof(resp)))) {
			mlx5_ib_dbg(dev, "copy to user failed\n");
			err = -EFAULT;
			goto err_core;
		}
	}

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

	return 0;

err_core:
	mlx5_cmd_destroy_srq(dev, &srq->msrq);

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

	return 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_cmd_arm_srq(dev, &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_cmd_query_srq(dev, &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 ib_udata *udata)
{
	struct mlx5_ib_dev *dev = to_mdev(srq->device);
	struct mlx5_ib_srq *msrq = to_msrq(srq);
	int ret;

	ret = mlx5_cmd_destroy_srq(dev, &msrq->msrq);
	if (ret)
		return ret;

	if (udata)
		destroy_srq_user(srq->pd, msrq, udata);
	else
		destroy_srq_kernel(dev, msrq);
	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, const struct ib_recv_wr *wr,
			  const 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
6cd0014a	1	// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
e126ba97	2	/*
6cd0014a	3	* Copyright (c) 2013-2018, Mellanox Technologies inc. All rights reserved.
e126ba97 EC	4	*/
	5
	6	#include <linux/module.h>
	7	#include <linux/mlx5/qp.h>
e126ba97 EC	8	#include <linux/slab.h>
e126ba97 EC	9	#include <rdma/ib_umem.h>
43bc8893	10	#include <rdma/ib_user_verbs.h>
e126ba97	11	#include "mlx5_ib.h"
f02d0d6e	12	#include "srq.h"
e126ba97 EC	13
	14	static void get_wqe(struct mlx5_ib_srq srq, int n)
	15	{
20e5a59b	16	return mlx5_frag_buf_get_wqe(&srq->fbc, n);
e126ba97 EC	17	}
	18
	19	static void mlx5_ib_srq_event(struct mlx5_core_srq *srq, enum mlx5_event type)
	20	{
	21	struct ib_event event;
	22	struct ib_srq *ibsrq = &to_mibsrq(srq)->ibsrq;
	23
	24	if (ibsrq->event_handler) {
	25	event.device = ibsrq->device;
	26	event.element.srq = ibsrq;
	27	switch (type) {
	28	case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
	29	event.event = IB_EVENT_SRQ_LIMIT_REACHED;
	30	break;
	31	case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
	32	event.event = IB_EVENT_SRQ_ERR;
	33	break;
	34	default:
	35	pr_warn("mlx5_ib: Unexpected event type %d on SRQ %06x\n",
	36	type, srq->srqn);
	37	return;
	38	}
	39
	40	ibsrq->event_handler(&event, ibsrq->srq_context);
	41	}
	42	}
	43
	44	static int create_srq_user(struct ib_pd pd, struct mlx5_ib_srq srq,
af1ba291 AK	45	struct mlx5_srq_attr *in,
af1ba291 AK	46	struct ib_udata *udata, int buf_size)
e126ba97 EC	47	{
e126ba97 EC	48	struct mlx5_ib_dev *dev = to_mdev(pd->device);
cfb5e088	49	struct mlx5_ib_create_srq ucmd = {};
89944450 SR	50	struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context(
89944450 SR	51	udata, struct mlx5_ib_ucontext, ibucontext);
43bc8893	52	size_t ucmdlen;
e126ba97	53	int err;
cfb5e088	54	u32 uidx = MLX5_IB_DEFAULT_UIDX;
e126ba97	55
3d943c9d	56	ucmdlen = min(udata->inlen, sizeof(ucmd));
43bc8893 YD	57
43bc8893 YD	58	if (ib_copy_from_udata(&ucmd, udata, ucmdlen)) {
e126ba97 EC	59	mlx5_ib_dbg(dev, "failed copy udata\n");
	60	return -EFAULT;
	61	}
43bc8893	62
cfb5e088	63	if (ucmd.reserved0 \|\| ucmd.reserved1)
43bc8893 YD	64	return -EINVAL;
43bc8893 YD	65
3d943c9d	66	if (udata->inlen > sizeof(ucmd) &&
cfb5e088	67	!ib_is_udata_cleared(udata, sizeof(ucmd),
3d943c9d	68	udata->inlen - sizeof(ucmd)))
cfb5e088 HA	69	return -EINVAL;
cfb5e088 HA	70
3fd3307e	71	if (in->type != IB_SRQT_BASIC) {
89944450	72	err = get_srq_user_index(ucontext, &ucmd, udata->inlen, &uidx);
85d9691c MD	73	if (err)
	74	return err;
	75	}
cfb5e088	76
e126ba97 EC	77	srq->wq_sig = !!(ucmd.flags & MLX5_SRQ_FLAG_SIGNATURE);
e126ba97 EC	78
e9eb90eb	79	srq->umem = ib_umem_get_peer(pd->device, ucmd.buf_addr, buf_size, 0, 0);
e126ba97 EC	80	if (IS_ERR(srq->umem)) {
	81	mlx5_ib_dbg(dev, "failed umem get, size %d\n", buf_size);
	82	err = PTR_ERR(srq->umem);
	83	return err;
	84	}
8a7904a6	85	in->umem = srq->umem;
e126ba97	86
89944450	87	err = mlx5_ib_db_map_user(ucontext, udata, ucmd.db_addr, &srq->db);
e126ba97 EC	88	if (err) {
e126ba97 EC	89	mlx5_ib_dbg(dev, "map doorbell failed\n");
8a7904a6	90	goto err_umem;
e126ba97 EC	91	}
e126ba97 EC	92
5aa3771d	93	in->uid = (in->type != IB_SRQT_XRC) ? to_mpd(pd)->uid : 0;
af1ba291	94	if (MLX5_CAP_GEN(dev->mdev, cqe_version) == MLX5_CQE_VERSION_V1 &&
3fd3307e	95	in->type != IB_SRQT_BASIC)
af1ba291	96	in->user_index = uidx;
cfb5e088	97
e126ba97 EC	98	return 0;
e126ba97 EC	99
e126ba97 EC	100	err_umem:
	101	ib_umem_release(srq->umem);
	102
	103	return err;
	104	}
	105
	106	static int create_srq_kernel(struct mlx5_ib_dev dev, struct mlx5_ib_srq srq,
af1ba291	107	struct mlx5_srq_attr *in, int buf_size)
e126ba97 EC	108	{
	109	int err;
	110	int i;
	111	struct mlx5_wqe_srq_next_seg *next;
e126ba97	112
9603b61d	113	err = mlx5_db_alloc(dev->mdev, &srq->db);
e126ba97 EC	114	if (err) {
	115	mlx5_ib_warn(dev, "alloc dbell rec failed\n");
	116	return err;
	117	}
	118
20e5a59b GL	119	if (mlx5_frag_buf_alloc_node(dev->mdev, buf_size, &srq->buf,
20e5a59b GL	120	dev->mdev->priv.numa_node)) {
e126ba97 EC	121	mlx5_ib_dbg(dev, "buf alloc failed\n");
	122	err = -ENOMEM;
	123	goto err_db;
	124	}
e126ba97	125
20e5a59b GL	126	mlx5_init_fbc(srq->buf.frags, srq->msrq.wqe_shift, ilog2(srq->msrq.max),
	127	&srq->fbc);
	128
e126ba97 EC	129	srq->head = 0;
	130	srq->tail = srq->msrq.max - 1;
	131	srq->wqe_ctr = 0;
	132
	133	for (i = 0; i < srq->msrq.max; i++) {
	134	next = get_wqe(srq, i);
	135	next->next_wqe_index =
	136	cpu_to_be16((i + 1) & (srq->msrq.max - 1));
	137	}
	138
0fd27a88	139	mlx5_ib_dbg(dev, "srq->buf.page_shift = %d\n", srq->buf.page_shift);
778e1cdd	140	in->pas = kvcalloc(srq->buf.npages, sizeof(*in->pas), GFP_KERNEL);
af1ba291	141	if (!in->pas) {
e126ba97 EC	142	err = -ENOMEM;
	143	goto err_buf;
	144	}
20e5a59b	145	mlx5_fill_page_frag_array(&srq->buf, in->pas);
e126ba97	146
b5883008	147	srq->wrid = kvmalloc_array(srq->msrq.max, sizeof(u64), GFP_KERNEL);
e126ba97	148	if (!srq->wrid) {
e126ba97 EC	149	err = -ENOMEM;
	150	goto err_in;
	151	}
c48d386b	152	srq->wq_sig = 0;
e126ba97	153
0fd27a88	154	in->log_page_size = srq->buf.page_shift - MLX5_ADAPTER_PAGE_SHIFT;
af1ba291	155	if (MLX5_CAP_GEN(dev->mdev, cqe_version) == MLX5_CQE_VERSION_V1 &&
3fd3307e	156	in->type != IB_SRQT_BASIC)
af1ba291	157	in->user_index = MLX5_IB_DEFAULT_UIDX;
cfb5e088	158
e126ba97 EC	159	return 0;
	160
	161	err_in:
af1ba291	162	kvfree(in->pas);
e126ba97 EC	163
e126ba97 EC	164	err_buf:
20e5a59b	165	mlx5_frag_buf_free(dev->mdev, &srq->buf);
e126ba97 EC	166
e126ba97 EC	167	err_db:
9603b61d	168	mlx5_db_free(dev->mdev, &srq->db);
e126ba97 EC	169	return err;
	170	}
	171
bdeacabd SR	172	static void destroy_srq_user(struct ib_pd pd, struct mlx5_ib_srq srq,
bdeacabd SR	173	struct ib_udata *udata)
e126ba97	174	{
bdeacabd SR	175	mlx5_ib_db_unmap_user(
	176	rdma_udata_to_drv_context(
	177	udata,
	178	struct mlx5_ib_ucontext,
	179	ibucontext),
	180	&srq->db);
e126ba97 EC	181	ib_umem_release(srq->umem);
	182	}
	183
	184
	185	static void destroy_srq_kernel(struct mlx5_ib_dev dev, struct mlx5_ib_srq srq)
	186	{
b5883008	187	kvfree(srq->wrid);
20e5a59b	188	mlx5_frag_buf_free(dev->mdev, &srq->buf);
9603b61d	189	mlx5_db_free(dev->mdev, &srq->db);
e126ba97 EC	190	}
e126ba97 EC	191
68e326de LR	192	int mlx5_ib_create_srq(struct ib_srq *ib_srq,
	193	struct ib_srq_init_attr *init_attr,
	194	struct ib_udata *udata)
e126ba97	195	{
68e326de LR	196	struct mlx5_ib_dev *dev = to_mdev(ib_srq->device);
68e326de LR	197	struct mlx5_ib_srq *srq = to_msrq(ib_srq);
c2b37f76 BP	198	size_t desc_size;
c2b37f76 BP	199	size_t buf_size;
e126ba97	200	int err;
68e326de	201	struct mlx5_srq_attr in = {};
938fe83c	202	__u32 max_srq_wqes = 1 << MLX5_CAP_GEN(dev->mdev, log_max_srq_sz);
e126ba97	203
652caba5 JG	204	if (init_attr->srq_type != IB_SRQT_BASIC &&
	205	init_attr->srq_type != IB_SRQT_XRC &&
	206	init_attr->srq_type != IB_SRQT_TM)
	207	return -EOPNOTSUPP;
	208
e126ba97	209	/* Sanity check SRQ size before proceeding */
938fe83c	210	if (init_attr->attr.max_wr >= max_srq_wqes) {
e126ba97 EC	211	mlx5_ib_dbg(dev, "max_wr %d, cap %d\n",
e126ba97 EC	212	init_attr->attr.max_wr,
938fe83c	213	max_srq_wqes);
68e326de	214	return -EINVAL;
e126ba97 EC	215	}
e126ba97 EC	216
e126ba97 EC	217	mutex_init(&srq->mutex);
	218	spin_lock_init(&srq->lock);
	219	srq->msrq.max = roundup_pow_of_two(init_attr->attr.max_wr + 1);
	220	srq->msrq.max_gs = init_attr->attr.max_sge;
	221
	222	desc_size = sizeof(struct mlx5_wqe_srq_next_seg) +
	223	srq->msrq.max_gs * sizeof(struct mlx5_wqe_data_seg);
68e326de LR	224	if (desc_size == 0 \|\| srq->msrq.max_gs > desc_size)
	225	return -EINVAL;
	226
e126ba97	227	desc_size = roundup_pow_of_two(desc_size);
c2b37f76	228	desc_size = max_t(size_t, 32, desc_size);
68e326de LR	229	if (desc_size < sizeof(struct mlx5_wqe_srq_next_seg))
	230	return -EINVAL;
	231
e126ba97 EC	232	srq->msrq.max_avail_gather = (desc_size - sizeof(struct mlx5_wqe_srq_next_seg)) /
	233	sizeof(struct mlx5_wqe_data_seg);
	234	srq->msrq.wqe_shift = ilog2(desc_size);
	235	buf_size = srq->msrq.max * desc_size;
68e326de LR	236	if (buf_size < desc_size)
	237	return -EINVAL;
	238
c73b7911	239	in.type = init_attr->srq_type;
e126ba97	240
e00b64f7	241	if (udata)
68e326de	242	err = create_srq_user(ib_srq->pd, srq, &in, udata, buf_size);
e126ba97	243	else
af1ba291	244	err = create_srq_kernel(dev, srq, &in, buf_size);
e126ba97 EC	245
	246	if (err) {
	247	mlx5_ib_warn(dev, "create srq %s failed, err %d\n",
e00b64f7	248	udata ? "user" : "kernel", err);
68e326de	249	return err;
e126ba97 EC	250	}
e126ba97 EC	251
af1ba291 AK	252	in.log_size = ilog2(srq->msrq.max);
	253	in.wqe_shift = srq->msrq.wqe_shift - 4;
	254	if (srq->wq_sig)
	255	in.flags \|= MLX5_SRQ_FLAG_WQ_SIG;
1a56ff6d	256
f4375443	257	if (init_attr->srq_type == IB_SRQT_XRC && init_attr->ext.xrc.xrcd)
af1ba291	258	in.xrcd = to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn;
1a56ff6d	259	else
f4375443	260	in.xrcd = dev->devr.xrcdn0;
1a56ff6d	261
3fd3307e AK	262	if (init_attr->srq_type == IB_SRQT_TM) {
	263	in.tm_log_list_size =
	264	ilog2(init_attr->ext.tag_matching.max_num_tags) + 1;
	265	if (in.tm_log_list_size >
	266	MLX5_CAP_GEN(dev->mdev, log_tag_matching_list_sz)) {
	267	mlx5_ib_dbg(dev, "TM SRQ max_num_tags exceeding limit\n");
	268	err = -EINVAL;
	269	goto err_usr_kern_srq;
	270	}
	271	in.flags \|= MLX5_SRQ_FLAG_RNDV;
	272	}
	273
1a56ff6d AK	274	if (ib_srq_has_cq(init_attr->srq_type))
	275	in.cqn = to_mcq(init_attr->ext.cq)->mcq.cqn;
	276	else
af1ba291	277	in.cqn = to_mcq(dev->devr.c0)->mcq.cqn;
e126ba97	278
68e326de	279	in.pd = to_mpd(ib_srq->pd)->pdn;
af1ba291	280	in.db_record = srq->db.dma;
b4990804	281	err = mlx5_cmd_create_srq(dev, &srq->msrq, &in);
af1ba291	282	kvfree(in.pas);
e126ba97 EC	283	if (err) {
e126ba97 EC	284	mlx5_ib_dbg(dev, "create SRQ failed, err %d\n", err);
56e1ab0f	285	goto err_usr_kern_srq;
e126ba97 EC	286	}
	287
	288	mlx5_ib_dbg(dev, "create SRQ with srqn 0x%x\n", srq->msrq.srqn);
	289
	290	srq->msrq.event = mlx5_ib_srq_event;
	291	srq->ibsrq.ext.xrc.srq_num = srq->msrq.srqn;
	292
cf26deff YH	293	if (udata) {
	294	struct mlx5_ib_create_srq_resp resp = {
	295	.srqn = srq->msrq.srqn,
	296	};
	297
	298	if (ib_copy_to_udata(udata, &resp, min(udata->outlen,
	299	sizeof(resp)))) {
e126ba97 EC	300	mlx5_ib_dbg(dev, "copy to user failed\n");
	301	err = -EFAULT;
	302	goto err_core;
	303	}
cf26deff	304	}
e126ba97 EC	305
	306	init_attr->attr.max_wr = srq->msrq.max - 1;
	307
68e326de	308	return 0;
e126ba97 EC	309
e126ba97 EC	310	err_core:
b4990804	311	mlx5_cmd_destroy_srq(dev, &srq->msrq);
56e1ab0f ML	312
56e1ab0f ML	313	err_usr_kern_srq:
e00b64f7	314	if (udata)
68e326de	315	destroy_srq_user(ib_srq->pd, srq, udata);
e126ba97 EC	316	else
	317	destroy_srq_kernel(dev, srq);
	318
68e326de	319	return err;
e126ba97 EC	320	}
	321
	322	int mlx5_ib_modify_srq(struct ib_srq ibsrq, struct ib_srq_attr attr,
	323	enum ib_srq_attr_mask attr_mask, struct ib_udata *udata)
	324	{
	325	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
	326	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
	327	int ret;
	328
	329	/* We don't support resizing SRQs yet */
	330	if (attr_mask & IB_SRQ_MAX_WR)
	331	return -EINVAL;
	332
	333	if (attr_mask & IB_SRQ_LIMIT) {
	334	if (attr->srq_limit >= srq->msrq.max)
	335	return -EINVAL;
	336
	337	mutex_lock(&srq->mutex);
b4990804	338	ret = mlx5_cmd_arm_srq(dev, &srq->msrq, attr->srq_limit, 1);
e126ba97 EC	339	mutex_unlock(&srq->mutex);
	340
	341	if (ret)
	342	return ret;
	343	}
	344
	345	return 0;
	346	}
	347
	348	int mlx5_ib_query_srq(struct ib_srq ibsrq, struct ib_srq_attr srq_attr)
	349	{
	350	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
	351	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
	352	int ret;
af1ba291	353	struct mlx5_srq_attr *out;
e126ba97 EC	354
	355	out = kzalloc(sizeof(*out), GFP_KERNEL);
	356	if (!out)
	357	return -ENOMEM;
	358
b4990804	359	ret = mlx5_cmd_query_srq(dev, &srq->msrq, out);
e126ba97 EC	360	if (ret)
	361	goto out_box;
	362
af1ba291	363	srq_attr->srq_limit = out->lwm;
e126ba97 EC	364	srq_attr->max_wr = srq->msrq.max - 1;
	365	srq_attr->max_sge = srq->msrq.max_gs;
	366
	367	out_box:
	368	kfree(out);
	369	return ret;
	370	}
	371
119181d1	372	int mlx5_ib_destroy_srq(struct ib_srq srq, struct ib_udata udata)
e126ba97 EC	373	{
	374	struct mlx5_ib_dev *dev = to_mdev(srq->device);
	375	struct mlx5_ib_srq *msrq = to_msrq(srq);
119181d1 LR	376	int ret;
	377
	378	ret = mlx5_cmd_destroy_srq(dev, &msrq->msrq);
	379	if (ret)
	380	return ret;
e126ba97	381
119181d1 LR	382	if (udata)
	383	destroy_srq_user(srq->pd, msrq, udata);
	384	else
1faacf82	385	destroy_srq_kernel(dev, msrq);
119181d1	386	return 0;
e126ba97 EC	387	}
	388
	389	void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index)
	390	{
	391	struct mlx5_wqe_srq_next_seg *next;
	392
	393	/* always called with interrupts disabled. */
	394	spin_lock(&srq->lock);
	395
	396	next = get_wqe(srq, srq->tail);
	397	next->next_wqe_index = cpu_to_be16(wqe_index);
	398	srq->tail = wqe_index;
	399
	400	spin_unlock(&srq->lock);
	401	}
	402
d34ac5cd BVA	403	int mlx5_ib_post_srq_recv(struct ib_srq ibsrq, const struct ib_recv_wr wr,
d34ac5cd BVA	404	const struct ib_recv_wr **bad_wr)
e126ba97 EC	405	{
	406	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
	407	struct mlx5_wqe_srq_next_seg *next;
	408	struct mlx5_wqe_data_seg *scat;
89ea94a7 MG	409	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
89ea94a7 MG	410	struct mlx5_core_dev *mdev = dev->mdev;
e126ba97 EC	411	unsigned long flags;
	412	int err = 0;
	413	int nreq;
	414	int i;
	415
	416	spin_lock_irqsave(&srq->lock, flags);
	417
89ea94a7 MG	418	if (mdev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
	419	err = -EIO;
	420	*bad_wr = wr;
	421	goto out;
	422	}
	423
e126ba97 EC	424	for (nreq = 0; wr; nreq++, wr = wr->next) {
	425	if (unlikely(wr->num_sge > srq->msrq.max_gs)) {
	426	err = -EINVAL;
	427	*bad_wr = wr;
	428	break;
	429	}
	430
	431	if (unlikely(srq->head == srq->tail)) {
	432	err = -ENOMEM;
	433	*bad_wr = wr;
	434	break;
	435	}
	436
	437	srq->wrid[srq->head] = wr->wr_id;
	438
	439	next = get_wqe(srq, srq->head);
	440	srq->head = be16_to_cpu(next->next_wqe_index);
	441	scat = (struct mlx5_wqe_data_seg *)(next + 1);
	442
	443	for (i = 0; i < wr->num_sge; i++) {
	444	scat[i].byte_count = cpu_to_be32(wr->sg_list[i].length);
	445	scat[i].lkey = cpu_to_be32(wr->sg_list[i].lkey);
	446	scat[i].addr = cpu_to_be64(wr->sg_list[i].addr);
	447	}
	448
	449	if (i < srq->msrq.max_avail_gather) {
	450	scat[i].byte_count = 0;
	451	scat[i].lkey = cpu_to_be32(MLX5_INVALID_LKEY);
	452	scat[i].addr = 0;
	453	}
	454	}
	455
	456	if (likely(nreq)) {
	457	srq->wqe_ctr += nreq;
	458
	459	/* Make sure that descriptors are written before
	460	* doorbell record.
	461	*/
	462	wmb();
	463
	464	*srq->db.db = cpu_to_be32(srq->wqe_ctr);
	465	}
89ea94a7	466	out:
e126ba97 EC	467	spin_unlock_irqrestore(&srq->lock, flags);
	468
	469	return err;
	470	}