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;
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);
if (err)
goto err_srq;
*srq->db.db = 0;
if (mlx4_buf_alloc(dev->dev, buf_size, PAGE_SIZE * 2, &srq->buf)) {
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));
}
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);
if (err)
goto err_mtt;
srq->wrid = kmalloc(srq->msrq.max * sizeof (u64), GFP_KERNEL);
if (!srq->wrid) {
err = -ENOMEM;
goto err_mtt;
}
}
err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, &srq->mtt,
srq->db.dma, &srq->msrq);
if (err)
goto err_wrid;
srq->msrq.event = mlx4_ib_srq_event;
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
kfree(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);
}
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_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;
}
in->pas = mlx5_vzalloc(sizeof(*in->pas) * ncont);
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_XRC)
in->user_index = uidx;
return 0;
err_in:
kvfree(in->pas);
err_umem:
ib_umem_release(srq->umem);
return err;
}
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)
{
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;
ucmdlen =
(udata->inlen - sizeof(struct ib_uverbs_cmd_hdr) <
sizeof(ucmd)) ? (sizeof(ucmd) -
sizeof(ucmd.reserved)) : sizeof(ucmd);
if (ib_copy_from_udata(&ucmd, udata, ucmdlen)) {
mlx5_ib_dbg(dev, "failed copy udata\n");
return -EFAULT;
}
if (ucmdlen == sizeof(ucmd) &&
ucmd.reserved != 0)
return -EINVAL;
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 - PAGE_SHIFT;
(*in)->ctx.pgoff_cqn = cpu_to_be32(offset << 26);
return 0;
err_in:
mlx5_vfree(*in);
err_umem:
ib_umem_release(srq->umem);
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);
}
struct ib_srq *mlx4_ib_create_xrc_srq(struct ib_pd *pd,
struct ib_cq *xrc_cq,
struct ib_xrcd *xrcd,
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;
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) {
mlx4_ib_dbg("a size param is out of range. "
"max_wr = 0x%x, max_sge = 0x%x",
init_attr->attr.max_wr, init_attr->attr.max_sge);
return ERR_PTR(-EINVAL);
}
srq = kzalloc(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 {
struct mlx4_wqe_data_seg *scatter;
err = mlx4_db_alloc(dev->dev, &srq->db, 0);
if (err)
goto err_srq;
*srq->db.db = 0;
if (mlx4_buf_alloc(dev->dev, buf_size, PAGE_SIZE * 2, &srq->buf)) {
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);
if (err)
goto err_mtt;
srq->wrid = kmalloc(srq->msrq.max * sizeof (u64), GFP_KERNEL);
if (!srq->wrid) {
err = -ENOMEM;
goto err_mtt;
}
}
cqn = xrc_cq ? (u32) (to_mcq(xrc_cq)->mcq.cqn) : 0;
xrcdn = xrcd ? (u16) (to_mxrcd(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;
if (pd->uobject) {
if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof (__u32))) {
err = -EFAULT;
goto err_wrid;
}
} else
srq->ibsrq.xrc_srq_num = srq->msrq.srqn;
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
kfree(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);
}
