[mirror_ubuntu-zesty-kernel.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"
#include "user.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_create_srq_mbox_in **in,
			   struct ib_udata *udata, int buf_size, int *inlen,
			   int is_xrc)
{
	struct mlx5_ib_dev *dev = to_mdev(pd->device);
	struct mlx5_ib_create_srq ucmd = {};
	size_t ucmdlen;
	void *xsrqc;
	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 (is_xrc) {
		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, &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;
	}

	*inlen = sizeof(**in) + sizeof(*(*in)->pas) * ncont;
	*in = mlx5_vzalloc(*inlen);
	if (!(*in)) {
		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)->ctx.log_pg_sz = page_shift - MLX5_ADAPTER_PAGE_SHIFT;
	(*in)->ctx.pgoff_cqn = cpu_to_be32(offset << 26);

	if ((MLX5_CAP_GEN(dev->mdev, cqe_version) == MLX5_CQE_VERSION_V1) &&
	     is_xrc){
		xsrqc = MLX5_ADDR_OF(create_xrc_srq_in, *in,
				     xrc_srq_context_entry);
		MLX5_SET(xrc_srqc, xsrqc, user_index, uidx);
	}

	return 0;

err_in:
	kvfree(*in);

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_create_srq_mbox_in **in, int buf_size,
			     int *inlen, int is_xrc)
{
	int err;
	int i;
	struct mlx5_wqe_srq_next_seg *next;
	int page_shift;
	int npages;
	void *xsrqc;

	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;
	}
	page_shift = srq->buf.page_shift;

	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));
	}

	npages = DIV_ROUND_UP(srq->buf.npages, 1 << (page_shift - PAGE_SHIFT));
	mlx5_ib_dbg(dev, "buf_size %d, page_shift %d, npages %d, calc npages %d\n",
		    buf_size, page_shift, srq->buf.npages, npages);
	*inlen = sizeof(**in) + sizeof(*(*in)->pas) * npages;
	*in = mlx5_vzalloc(*inlen);
	if (!*in) {
		err = -ENOMEM;
		goto err_buf;
	}
	mlx5_fill_page_array(&srq->buf, (*in)->pas);

	srq->wrid = kmalloc(srq->msrq.max * sizeof(u64), GFP_KERNEL);
	if (!srq->wrid) {
		mlx5_ib_dbg(dev, "kmalloc failed %lu\n",
			    (unsigned long)(srq->msrq.max * sizeof(u64)));
		err = -ENOMEM;
		goto err_in;
	}
	srq->wq_sig = !!srq_signature;

	(*in)->ctx.log_pg_sz = page_shift - MLX5_ADAPTER_PAGE_SHIFT;

	if ((MLX5_CAP_GEN(dev->mdev, cqe_version) == MLX5_CQE_VERSION_V1) &&
	     is_xrc){
		xsrqc = MLX5_ADDR_OF(create_xrc_srq_in, *in,
				     xrc_srq_context_entry);
		/* 0xffffff means we ask to work with cqe version 0 */
		MLX5_SET(xrc_srqc, xsrqc, user_index, MLX5_IB_DEFAULT_UIDX);
	}

	return 0;

err_in:
	kvfree(*in);

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)
{
	kfree(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;
	int desc_size;
	int buf_size;
	int err;
	struct mlx5_create_srq_mbox_in *uninitialized_var(in);
	int uninitialized_var(inlen);
	int is_xrc;
	u32 flgs, xrcdn;
	__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);
	desc_size = roundup_pow_of_two(desc_size);
	desc_size = max_t(int, 32, desc_size);
	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;
	mlx5_ib_dbg(dev, "desc_size 0x%x, req wr 0x%x, srq size 0x%x, max_gs 0x%x, max_avail_gather 0x%x\n",
		    desc_size, init_attr->attr.max_wr, srq->msrq.max, srq->msrq.max_gs,
		    srq->msrq.max_avail_gather);

	is_xrc = (init_attr->srq_type == IB_SRQT_XRC);

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

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

	in->ctx.state_log_sz = ilog2(srq->msrq.max);
	flgs = ((srq->msrq.wqe_shift - 4) | (is_xrc << 5) | (srq->wq_sig << 7)) << 24;
	xrcdn = 0;
	if (is_xrc) {
		xrcdn = to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn;
		in->ctx.pgoff_cqn |= cpu_to_be32(to_mcq(init_attr->ext.xrc.cq)->mcq.cqn);
	} else if (init_attr->srq_type == IB_SRQT_BASIC) {
		xrcdn = to_mxrcd(dev->devr.x0)->xrcdn;
		in->ctx.pgoff_cqn |= cpu_to_be32(to_mcq(dev->devr.c0)->mcq.cqn);
	}

	in->ctx.flags_xrcd = cpu_to_be32((flgs & 0xFF000000) | (xrcdn & 0xFFFFFF));

	in->ctx.pd = cpu_to_be32(to_mpd(pd)->pdn);
	in->ctx.db_record = cpu_to_be64(srq->db.dma);
	err = mlx5_core_create_srq(dev->mdev, &srq->msrq, in, inlen, is_xrc);
	kvfree(in);
	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_query_srq_mbox_out *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 = be16_to_cpu(out->ctx.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;
	unsigned long flags;
	int err = 0;
	int nreq;
	int i;

	spin_lock_irqsave(&srq->lock, flags);

	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);
	}

	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
	40	#include "mlx5_ib.h"
	41	#include "user.h"
	42
	43	/* not supported currently */
	44	static int srq_signature;
	45
	46	static void get_wqe(struct mlx5_ib_srq srq, int n)
	47	{
	48	return mlx5_buf_offset(&srq->buf, n << srq->msrq.wqe_shift);
	49	}
	50
	51	static void mlx5_ib_srq_event(struct mlx5_core_srq *srq, enum mlx5_event type)
	52	{
	53	struct ib_event event;
	54	struct ib_srq *ibsrq = &to_mibsrq(srq)->ibsrq;
	55
	56	if (ibsrq->event_handler) {
	57	event.device = ibsrq->device;
	58	event.element.srq = ibsrq;
	59	switch (type) {
	60	case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
	61	event.event = IB_EVENT_SRQ_LIMIT_REACHED;
	62	break;
	63	case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
	64	event.event = IB_EVENT_SRQ_ERR;
	65	break;
	66	default:
	67	pr_warn("mlx5_ib: Unexpected event type %d on SRQ %06x\n",
	68	type, srq->srqn);
	69	return;
	70	}
	71
	72	ibsrq->event_handler(&event, ibsrq->srq_context);
	73	}
	74	}
	75
	76	static int create_srq_user(struct ib_pd pd, struct mlx5_ib_srq srq,
	77	struct mlx5_create_srq_mbox_in **in,
85d9691c MD	78	struct ib_udata udata, int buf_size, int inlen,
85d9691c MD	79	int is_xrc)
e126ba97 EC	80	{
e126ba97 EC	81	struct mlx5_ib_dev *dev = to_mdev(pd->device);
cfb5e088	82	struct mlx5_ib_create_srq ucmd = {};
43bc8893	83	size_t ucmdlen;
cfb5e088	84	void *xsrqc;
e126ba97 EC	85	int err;
	86	int npages;
	87	int page_shift;
	88	int ncont;
	89	u32 offset;
cfb5e088	90	u32 uidx = MLX5_IB_DEFAULT_UIDX;
e126ba97	91
3d943c9d	92	ucmdlen = min(udata->inlen, sizeof(ucmd));
43bc8893 YD	93
43bc8893 YD	94	if (ib_copy_from_udata(&ucmd, udata, ucmdlen)) {
e126ba97 EC	95	mlx5_ib_dbg(dev, "failed copy udata\n");
	96	return -EFAULT;
	97	}
43bc8893	98
cfb5e088	99	if (ucmd.reserved0 \|\| ucmd.reserved1)
43bc8893 YD	100	return -EINVAL;
43bc8893 YD	101
3d943c9d	102	if (udata->inlen > sizeof(ucmd) &&
cfb5e088	103	!ib_is_udata_cleared(udata, sizeof(ucmd),
3d943c9d	104	udata->inlen - sizeof(ucmd)))
cfb5e088 HA	105	return -EINVAL;
cfb5e088 HA	106
85d9691c MD	107	if (is_xrc) {
	108	err = get_srq_user_index(to_mucontext(pd->uobject->context),
	109	&ucmd, udata->inlen, &uidx);
	110	if (err)
	111	return err;
	112	}
cfb5e088	113
e126ba97 EC	114	srq->wq_sig = !!(ucmd.flags & MLX5_SRQ_FLAG_SIGNATURE);
	115
	116	srq->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr, buf_size,
	117	0, 0);
	118	if (IS_ERR(srq->umem)) {
	119	mlx5_ib_dbg(dev, "failed umem get, size %d\n", buf_size);
	120	err = PTR_ERR(srq->umem);
	121	return err;
	122	}
	123
	124	mlx5_ib_cont_pages(srq->umem, ucmd.buf_addr, &npages,
	125	&page_shift, &ncont, NULL);
	126	err = mlx5_ib_get_buf_offset(ucmd.buf_addr, page_shift,
	127	&offset);
	128	if (err) {
	129	mlx5_ib_warn(dev, "bad offset\n");
	130	goto err_umem;
	131	}
	132
	133	inlen = sizeof(in) + sizeof((in)->pas) ncont;
	134	in = mlx5_vzalloc(inlen);
	135	if (!(*in)) {
	136	err = -ENOMEM;
	137	goto err_umem;
	138	}
	139
	140	mlx5_ib_populate_pas(dev, srq->umem, page_shift, (*in)->pas, 0);
	141
	142	err = mlx5_ib_db_map_user(to_mucontext(pd->uobject->context),
	143	ucmd.db_addr, &srq->db);
	144	if (err) {
	145	mlx5_ib_dbg(dev, "map doorbell failed\n");
	146	goto err_in;
	147	}
	148
1b77d2bd	149	(*in)->ctx.log_pg_sz = page_shift - MLX5_ADAPTER_PAGE_SHIFT;
e126ba97 EC	150	(*in)->ctx.pgoff_cqn = cpu_to_be32(offset << 26);
e126ba97 EC	151
85d9691c MD	152	if ((MLX5_CAP_GEN(dev->mdev, cqe_version) == MLX5_CQE_VERSION_V1) &&
85d9691c MD	153	is_xrc){
cfb5e088 HA	154	xsrqc = MLX5_ADDR_OF(create_xrc_srq_in, *in,
	155	xrc_srq_context_entry);
	156	MLX5_SET(xrc_srqc, xsrqc, user_index, uidx);
	157	}
	158
e126ba97 EC	159	return 0;
	160
	161	err_in:
479163f4	162	kvfree(*in);
e126ba97 EC	163
	164	err_umem:
	165	ib_umem_release(srq->umem);
	166
	167	return err;
	168	}
	169
	170	static int create_srq_kernel(struct mlx5_ib_dev dev, struct mlx5_ib_srq srq,
	171	struct mlx5_create_srq_mbox_in **in, int buf_size,
85d9691c	172	int *inlen, int is_xrc)
e126ba97 EC	173	{
	174	int err;
	175	int i;
	176	struct mlx5_wqe_srq_next_seg *next;
	177	int page_shift;
	178	int npages;
cfb5e088	179	void *xsrqc;
e126ba97	180
9603b61d	181	err = mlx5_db_alloc(dev->mdev, &srq->db);
e126ba97 EC	182	if (err) {
	183	mlx5_ib_warn(dev, "alloc dbell rec failed\n");
	184	return err;
	185	}
	186
64ffaa21	187	if (mlx5_buf_alloc(dev->mdev, buf_size, &srq->buf)) {
e126ba97 EC	188	mlx5_ib_dbg(dev, "buf alloc failed\n");
	189	err = -ENOMEM;
	190	goto err_db;
	191	}
	192	page_shift = srq->buf.page_shift;
	193
	194	srq->head = 0;
	195	srq->tail = srq->msrq.max - 1;
	196	srq->wqe_ctr = 0;
	197
	198	for (i = 0; i < srq->msrq.max; i++) {
	199	next = get_wqe(srq, i);
	200	next->next_wqe_index =
	201	cpu_to_be16((i + 1) & (srq->msrq.max - 1));
	202	}
	203
	204	npages = DIV_ROUND_UP(srq->buf.npages, 1 << (page_shift - PAGE_SHIFT));
	205	mlx5_ib_dbg(dev, "buf_size %d, page_shift %d, npages %d, calc npages %d\n",
	206	buf_size, page_shift, srq->buf.npages, npages);
	207	inlen = sizeof(in) + sizeof((in)->pas) npages;
	208	in = mlx5_vzalloc(inlen);
	209	if (!*in) {
	210	err = -ENOMEM;
	211	goto err_buf;
	212	}
	213	mlx5_fill_page_array(&srq->buf, (*in)->pas);
	214
	215	srq->wrid = kmalloc(srq->msrq.max * sizeof(u64), GFP_KERNEL);
	216	if (!srq->wrid) {
	217	mlx5_ib_dbg(dev, "kmalloc failed %lu\n",
	218	(unsigned long)(srq->msrq.max * sizeof(u64)));
	219	err = -ENOMEM;
	220	goto err_in;
	221	}
	222	srq->wq_sig = !!srq_signature;
	223
1b77d2bd	224	(*in)->ctx.log_pg_sz = page_shift - MLX5_ADAPTER_PAGE_SHIFT;
e126ba97	225
85d9691c MD	226	if ((MLX5_CAP_GEN(dev->mdev, cqe_version) == MLX5_CQE_VERSION_V1) &&
85d9691c MD	227	is_xrc){
cfb5e088 HA	228	xsrqc = MLX5_ADDR_OF(create_xrc_srq_in, *in,
	229	xrc_srq_context_entry);
	230	/* 0xffffff means we ask to work with cqe version 0 */
	231	MLX5_SET(xrc_srqc, xsrqc, user_index, MLX5_IB_DEFAULT_UIDX);
	232	}
	233
e126ba97 EC	234	return 0;
	235
	236	err_in:
479163f4	237	kvfree(*in);
e126ba97 EC	238
e126ba97 EC	239	err_buf:
9603b61d	240	mlx5_buf_free(dev->mdev, &srq->buf);
e126ba97 EC	241
e126ba97 EC	242	err_db:
9603b61d	243	mlx5_db_free(dev->mdev, &srq->db);
e126ba97 EC	244	return err;
	245	}
	246
	247	static void destroy_srq_user(struct ib_pd pd, struct mlx5_ib_srq srq)
	248	{
	249	mlx5_ib_db_unmap_user(to_mucontext(pd->uobject->context), &srq->db);
	250	ib_umem_release(srq->umem);
	251	}
	252
	253
	254	static void destroy_srq_kernel(struct mlx5_ib_dev dev, struct mlx5_ib_srq srq)
	255	{
	256	kfree(srq->wrid);
9603b61d JM	257	mlx5_buf_free(dev->mdev, &srq->buf);
9603b61d JM	258	mlx5_db_free(dev->mdev, &srq->db);
e126ba97 EC	259	}
	260
	261	struct ib_srq mlx5_ib_create_srq(struct ib_pd pd,
	262	struct ib_srq_init_attr *init_attr,
	263	struct ib_udata *udata)
	264	{
	265	struct mlx5_ib_dev *dev = to_mdev(pd->device);
	266	struct mlx5_ib_srq *srq;
	267	int desc_size;
	268	int buf_size;
	269	int err;
	270	struct mlx5_create_srq_mbox_in *uninitialized_var(in);
	271	int uninitialized_var(inlen);
	272	int is_xrc;
	273	u32 flgs, xrcdn;
938fe83c	274	__u32 max_srq_wqes = 1 << MLX5_CAP_GEN(dev->mdev, log_max_srq_sz);
e126ba97 EC	275
e126ba97 EC	276	/* Sanity check SRQ size before proceeding */
938fe83c	277	if (init_attr->attr.max_wr >= max_srq_wqes) {
e126ba97 EC	278	mlx5_ib_dbg(dev, "max_wr %d, cap %d\n",
e126ba97 EC	279	init_attr->attr.max_wr,
938fe83c	280	max_srq_wqes);
e126ba97 EC	281	return ERR_PTR(-EINVAL);
	282	}
	283
	284	srq = kmalloc(sizeof(*srq), GFP_KERNEL);
	285	if (!srq)
	286	return ERR_PTR(-ENOMEM);
	287
	288	mutex_init(&srq->mutex);
	289	spin_lock_init(&srq->lock);
	290	srq->msrq.max = roundup_pow_of_two(init_attr->attr.max_wr + 1);
	291	srq->msrq.max_gs = init_attr->attr.max_sge;
	292
	293	desc_size = sizeof(struct mlx5_wqe_srq_next_seg) +
	294	srq->msrq.max_gs * sizeof(struct mlx5_wqe_data_seg);
	295	desc_size = roundup_pow_of_two(desc_size);
	296	desc_size = max_t(int, 32, desc_size);
	297	srq->msrq.max_avail_gather = (desc_size - sizeof(struct mlx5_wqe_srq_next_seg)) /
	298	sizeof(struct mlx5_wqe_data_seg);
	299	srq->msrq.wqe_shift = ilog2(desc_size);
	300	buf_size = srq->msrq.max * desc_size;
	301	mlx5_ib_dbg(dev, "desc_size 0x%x, req wr 0x%x, srq size 0x%x, max_gs 0x%x, max_avail_gather 0x%x\n",
	302	desc_size, init_attr->attr.max_wr, srq->msrq.max, srq->msrq.max_gs,
	303	srq->msrq.max_avail_gather);
	304
85d9691c MD	305	is_xrc = (init_attr->srq_type == IB_SRQT_XRC);
85d9691c MD	306
e126ba97	307	if (pd->uobject)
85d9691c MD	308	err = create_srq_user(pd, srq, &in, udata, buf_size, &inlen,
85d9691c MD	309	is_xrc);
e126ba97	310	else
85d9691c MD	311	err = create_srq_kernel(dev, srq, &in, buf_size, &inlen,
85d9691c MD	312	is_xrc);
e126ba97 EC	313
	314	if (err) {
	315	mlx5_ib_warn(dev, "create srq %s failed, err %d\n",
	316	pd->uobject ? "user" : "kernel", err);
	317	goto err_srq;
	318	}
	319
e126ba97 EC	320	in->ctx.state_log_sz = ilog2(srq->msrq.max);
	321	flgs = ((srq->msrq.wqe_shift - 4) \| (is_xrc << 5) \| (srq->wq_sig << 7)) << 24;
	322	xrcdn = 0;
	323	if (is_xrc) {
	324	xrcdn = to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn;
	325	in->ctx.pgoff_cqn \|= cpu_to_be32(to_mcq(init_attr->ext.xrc.cq)->mcq.cqn);
	326	} else if (init_attr->srq_type == IB_SRQT_BASIC) {
	327	xrcdn = to_mxrcd(dev->devr.x0)->xrcdn;
	328	in->ctx.pgoff_cqn \|= cpu_to_be32(to_mcq(dev->devr.c0)->mcq.cqn);
	329	}
	330
	331	in->ctx.flags_xrcd = cpu_to_be32((flgs & 0xFF000000) \| (xrcdn & 0xFFFFFF));
	332
	333	in->ctx.pd = cpu_to_be32(to_mpd(pd)->pdn);
	334	in->ctx.db_record = cpu_to_be64(srq->db.dma);
01949d01	335	err = mlx5_core_create_srq(dev->mdev, &srq->msrq, in, inlen, is_xrc);
479163f4	336	kvfree(in);
e126ba97 EC	337	if (err) {
e126ba97 EC	338	mlx5_ib_dbg(dev, "create SRQ failed, err %d\n", err);
56e1ab0f	339	goto err_usr_kern_srq;
e126ba97 EC	340	}
	341
	342	mlx5_ib_dbg(dev, "create SRQ with srqn 0x%x\n", srq->msrq.srqn);
	343
	344	srq->msrq.event = mlx5_ib_srq_event;
	345	srq->ibsrq.ext.xrc.srq_num = srq->msrq.srqn;
	346
	347	if (pd->uobject)
	348	if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof(__u32))) {
	349	mlx5_ib_dbg(dev, "copy to user failed\n");
	350	err = -EFAULT;
	351	goto err_core;
	352	}
	353
	354	init_attr->attr.max_wr = srq->msrq.max - 1;
	355
	356	return &srq->ibsrq;
	357
	358	err_core:
9603b61d	359	mlx5_core_destroy_srq(dev->mdev, &srq->msrq);
56e1ab0f ML	360
56e1ab0f ML	361	err_usr_kern_srq:
e126ba97 EC	362	if (pd->uobject)
	363	destroy_srq_user(pd, srq);
	364	else
	365	destroy_srq_kernel(dev, srq);
	366
	367	err_srq:
	368	kfree(srq);
	369
	370	return ERR_PTR(err);
	371	}
	372
	373	int mlx5_ib_modify_srq(struct ib_srq ibsrq, struct ib_srq_attr attr,
	374	enum ib_srq_attr_mask attr_mask, struct ib_udata *udata)
	375	{
	376	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
	377	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
	378	int ret;
	379
	380	/* We don't support resizing SRQs yet */
	381	if (attr_mask & IB_SRQ_MAX_WR)
	382	return -EINVAL;
	383
	384	if (attr_mask & IB_SRQ_LIMIT) {
	385	if (attr->srq_limit >= srq->msrq.max)
	386	return -EINVAL;
	387
	388	mutex_lock(&srq->mutex);
9603b61d	389	ret = mlx5_core_arm_srq(dev->mdev, &srq->msrq, attr->srq_limit, 1);
e126ba97 EC	390	mutex_unlock(&srq->mutex);
	391
	392	if (ret)
	393	return ret;
	394	}
	395
	396	return 0;
	397	}
	398
	399	int mlx5_ib_query_srq(struct ib_srq ibsrq, struct ib_srq_attr srq_attr)
	400	{
	401	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
	402	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
	403	int ret;
	404	struct mlx5_query_srq_mbox_out *out;
	405
	406	out = kzalloc(sizeof(*out), GFP_KERNEL);
	407	if (!out)
	408	return -ENOMEM;
	409
9603b61d	410	ret = mlx5_core_query_srq(dev->mdev, &srq->msrq, out);
e126ba97 EC	411	if (ret)
	412	goto out_box;
	413
	414	srq_attr->srq_limit = be16_to_cpu(out->ctx.lwm);
	415	srq_attr->max_wr = srq->msrq.max - 1;
	416	srq_attr->max_sge = srq->msrq.max_gs;
	417
	418	out_box:
	419	kfree(out);
	420	return ret;
	421	}
	422
	423	int mlx5_ib_destroy_srq(struct ib_srq *srq)
	424	{
	425	struct mlx5_ib_dev *dev = to_mdev(srq->device);
	426	struct mlx5_ib_srq *msrq = to_msrq(srq);
	427
9603b61d	428	mlx5_core_destroy_srq(dev->mdev, &msrq->msrq);
e126ba97 EC	429
	430	if (srq->uobject) {
	431	mlx5_ib_db_unmap_user(to_mucontext(srq->uobject->context), &msrq->db);
	432	ib_umem_release(msrq->umem);
	433	} else {
1faacf82	434	destroy_srq_kernel(dev, msrq);
e126ba97 EC	435	}
	436
	437	kfree(srq);
	438	return 0;
	439	}
	440
	441	void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index)
	442	{
	443	struct mlx5_wqe_srq_next_seg *next;
	444
	445	/* always called with interrupts disabled. */
	446	spin_lock(&srq->lock);
	447
	448	next = get_wqe(srq, srq->tail);
	449	next->next_wqe_index = cpu_to_be16(wqe_index);
	450	srq->tail = wqe_index;
	451
	452	spin_unlock(&srq->lock);
	453	}
	454
	455	int mlx5_ib_post_srq_recv(struct ib_srq ibsrq, struct ib_recv_wr wr,
	456	struct ib_recv_wr **bad_wr)
	457	{
	458	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
	459	struct mlx5_wqe_srq_next_seg *next;
	460	struct mlx5_wqe_data_seg *scat;
	461	unsigned long flags;
	462	int err = 0;
	463	int nreq;
	464	int i;
	465
	466	spin_lock_irqsave(&srq->lock, flags);
	467
	468	for (nreq = 0; wr; nreq++, wr = wr->next) {
	469	if (unlikely(wr->num_sge > srq->msrq.max_gs)) {
	470	err = -EINVAL;
	471	*bad_wr = wr;
	472	break;
	473	}
	474
	475	if (unlikely(srq->head == srq->tail)) {
	476	err = -ENOMEM;
	477	*bad_wr = wr;
	478	break;
	479	}
	480
	481	srq->wrid[srq->head] = wr->wr_id;
	482
	483	next = get_wqe(srq, srq->head);
	484	srq->head = be16_to_cpu(next->next_wqe_index);
	485	scat = (struct mlx5_wqe_data_seg *)(next + 1);
	486
	487	for (i = 0; i < wr->num_sge; i++) {
	488	scat[i].byte_count = cpu_to_be32(wr->sg_list[i].length);
	489	scat[i].lkey = cpu_to_be32(wr->sg_list[i].lkey);
	490	scat[i].addr = cpu_to_be64(wr->sg_list[i].addr);
	491	}
	492
	493	if (i < srq->msrq.max_avail_gather) {
	494	scat[i].byte_count = 0;
	495	scat[i].lkey = cpu_to_be32(MLX5_INVALID_LKEY);
	496	scat[i].addr = 0;
	497	}
	498	}
499
500	if (likely(nreq)) {
501	srq->wqe_ctr += nreq;
502
503	/* Make sure that descriptors are written before
504	* doorbell record.
505	*/
506	wmb();
507
508	*srq->db.db = cpu_to_be32(srq->wqe_ctr);
509	}
510
511	spin_unlock_irqrestore(&srq->lock, flags);
512
513	return err;
514	}