int mlx4_alloc_hwq_res(struct mlx4_dev *dev, struct mlx4_hwq_resources *wqres,
int size, int max_direct)
{
int err;
err = mlx4_db_alloc(dev, &wqres->db, 1);
if (err)
return err;
*wqres->db.db = 0;
err = mlx4_buf_alloc(dev, size, max_direct, &wqres->buf);
if (err)
goto err_db;
err = mlx4_mtt_init(dev, wqres->buf.npages, wqres->buf.page_shift,
&wqres->mtt);
if (err)
goto err_buf;
err = mlx4_buf_write_mtt(dev, &wqres->mtt, &wqres->buf);
if (err)
goto err_mtt;
return 0;
err_mtt:
mlx4_mtt_cleanup(dev, &wqres->mtt);
err_buf:
mlx4_buf_free(dev, size, &wqres->buf);
err_db:
mlx4_db_free(dev, &wqres->db);
return err;
}
struct ib_cq *mlx4_ib_create_cq(struct ib_device *ibdev, int entries, int vector,
struct ib_ucontext *context,
struct ib_udata *udata)
{
struct mlx4_ib_dev *dev = to_mdev(ibdev);
struct mlx4_ib_cq *cq;
struct mlx4_uar *uar;
int err;
if (entries < 1 || entries > dev->dev->caps.max_cqes)
return ERR_PTR(-EINVAL);
cq = kmalloc(sizeof *cq, GFP_KERNEL);
if (!cq)
return ERR_PTR(-ENOMEM);
entries = roundup_pow_of_two(entries + 1);
cq->ibcq.cqe = entries - 1;
mutex_init(&cq->resize_mutex);
spin_lock_init(&cq->lock);
cq->resize_buf = NULL;
cq->resize_umem = NULL;
if (context) {
struct mlx4_ib_create_cq ucmd;
if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
err = -EFAULT;
goto err_cq;
}
err = mlx4_ib_get_cq_umem(dev, context, &cq->buf, &cq->umem,
ucmd.buf_addr, entries);
if (err)
goto err_cq;
err = mlx4_ib_db_map_user(to_mucontext(context), ucmd.db_addr,
&cq->db);
if (err)
goto err_mtt;
uar = &to_mucontext(context)->uar;
} else {
err = mlx4_db_alloc(dev->dev, &cq->db, 1);
if (err)
goto err_cq;
cq->mcq.set_ci_db = cq->db.db;
cq->mcq.arm_db = cq->db.db + 1;
*cq->mcq.set_ci_db = 0;
*cq->mcq.arm_db = 0;
err = mlx4_ib_alloc_cq_buf(dev, &cq->buf, entries);
if (err)
goto err_db;
uar = &dev->priv_uar;
}
err = mlx4_cq_alloc(dev->dev, entries, &cq->buf.mtt, uar,
cq->db.dma, &cq->mcq, vector, 0);
if (err)
goto err_dbmap;
cq->mcq.comp = mlx4_ib_cq_comp;
cq->mcq.event = mlx4_ib_cq_event;
if (context)
if (ib_copy_to_udata(udata, &cq->mcq.cqn, sizeof (__u32))) {
err = -EFAULT;
goto err_dbmap;
}
return &cq->ibcq;
err_dbmap:
if (context)
mlx4_ib_db_unmap_user(to_mucontext(context), &cq->db);
err_mtt:
mlx4_mtt_cleanup(dev->dev, &cq->buf.mtt);
if (context)
ib_umem_release(cq->umem);
else
mlx4_ib_free_cq_buf(dev, &cq->buf, cq->ibcq.cqe);
err_db:
if (!context)
mlx4_db_free(dev->dev, &cq->db);
err_cq:
kfree(cq);
return ERR_PTR(err);
}
struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
struct ib_srq_init_attr *init_attr,
struct ib_udata *udata)
{
struct mlx4_ib_dev *dev = to_mdev(pd->device);
struct mlx4_ib_srq *srq;
struct mlx4_wqe_srq_next_seg *next;
struct mlx4_wqe_data_seg *scatter;
u32 cqn;
u16 xrcdn;
int desc_size;
int buf_size;
int err;
int i;
/* Sanity check SRQ size before proceeding */
if (init_attr->attr.max_wr >= dev->dev->caps.max_srq_wqes ||
init_attr->attr.max_sge > dev->dev->caps.max_srq_sge)
return ERR_PTR(-EINVAL);
srq = kmalloc(sizeof *srq, GFP_KERNEL);
if (!srq)
return ERR_PTR(-ENOMEM);
mutex_init(&srq->mutex);
spin_lock_init(&srq->lock);
srq->msrq.max = roundup_pow_of_two(init_attr->attr.max_wr + 1);
srq->msrq.max_gs = init_attr->attr.max_sge;
desc_size = max(32UL,
roundup_pow_of_two(sizeof (struct mlx4_wqe_srq_next_seg) +
srq->msrq.max_gs *
sizeof (struct mlx4_wqe_data_seg)));
srq->msrq.wqe_shift = ilog2(desc_size);
buf_size = srq->msrq.max * desc_size;
if (pd->uobject) {
struct mlx4_ib_create_srq ucmd;
if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
err = -EFAULT;
goto err_srq;
}
srq->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr,
buf_size, 0, 0);
if (IS_ERR(srq->umem)) {
err = PTR_ERR(srq->umem);
goto err_srq;
}
err = mlx4_mtt_init(dev->dev, ib_umem_page_count(srq->umem),
ilog2(srq->umem->page_size), &srq->mtt);
if (err)
goto err_buf;
err = mlx4_ib_umem_write_mtt(dev, &srq->mtt, srq->umem);
if (err)
goto err_mtt;
err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context),
ucmd.db_addr, &srq->db);
if (err)
goto err_mtt;
} else {
err = mlx4_db_alloc(dev->dev, &srq->db, 0, GFP_KERNEL);
if (err)
goto err_srq;
*srq->db.db = 0;
if (mlx4_buf_alloc(dev->dev, buf_size, PAGE_SIZE * 2, &srq->buf,
GFP_KERNEL)) {
err = -ENOMEM;
goto err_db;
}
srq->head = 0;
srq->tail = srq->msrq.max - 1;
srq->wqe_ctr = 0;
for (i = 0; i < srq->msrq.max; ++i) {
next = get_wqe(srq, i);
next->next_wqe_index =
cpu_to_be16((i + 1) & (srq->msrq.max - 1));
for (scatter = (void *) (next + 1);
(void *) scatter < (void *) next + desc_size;
++scatter)
scatter->lkey = cpu_to_be32(MLX4_INVALID_LKEY);
}
err = mlx4_mtt_init(dev->dev, srq->buf.npages, srq->buf.page_shift,
&srq->mtt);
if (err)
goto err_buf;
err = mlx4_buf_write_mtt(dev->dev, &srq->mtt, &srq->buf, GFP_KERNEL);
if (err)
goto err_mtt;
srq->wrid = kmalloc(srq->msrq.max * sizeof (u64), GFP_KERNEL);
if (!srq->wrid) {
err = -ENOMEM;
goto err_mtt;
}
}
cqn = (init_attr->srq_type == IB_SRQT_XRC) ?
to_mcq(init_attr->ext.xrc.cq)->mcq.cqn : 0;
xrcdn = (init_attr->srq_type == IB_SRQT_XRC) ?
to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn :
(u16) dev->dev->caps.reserved_xrcds;
err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, cqn, xrcdn, &srq->mtt,
srq->db.dma, &srq->msrq);
if (err)
goto err_wrid;
srq->msrq.event = mlx4_ib_srq_event;
srq->ibsrq.ext.xrc.srq_num = srq->msrq.srqn;
if (pd->uobject)
if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof (__u32))) {
err = -EFAULT;
goto err_wrid;
}
init_attr->attr.max_wr = srq->msrq.max - 1;
return &srq->ibsrq;
err_wrid:
if (pd->uobject)
mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context), &srq->db);
else
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);
}
struct ib_cq *mlx4_ib_create_cq(struct ib_device *ibdev,
const struct ib_cq_init_attr *attr,
struct ib_udata *udata)
{
int entries = attr->cqe;
int vector = attr->comp_vector;
struct mlx4_ib_dev *dev = to_mdev(ibdev);
struct mlx4_ib_cq *cq;
struct mlx4_uar *uar;
void *buf_addr;
int err;
struct mlx4_ib_ucontext *context = rdma_udata_to_drv_context(
udata, struct mlx4_ib_ucontext, ibucontext);
if (entries < 1 || entries > dev->dev->caps.max_cqes)
return ERR_PTR(-EINVAL);
if (attr->flags & ~CQ_CREATE_FLAGS_SUPPORTED)
return ERR_PTR(-EINVAL);
cq = kzalloc(sizeof(*cq), GFP_KERNEL);
if (!cq)
return ERR_PTR(-ENOMEM);
entries = roundup_pow_of_two(entries + 1);
cq->ibcq.cqe = entries - 1;
mutex_init(&cq->resize_mutex);
spin_lock_init(&cq->lock);
cq->resize_buf = NULL;
cq->resize_umem = NULL;
cq->create_flags = attr->flags;
INIT_LIST_HEAD(&cq->send_qp_list);
INIT_LIST_HEAD(&cq->recv_qp_list);
if (udata) {
struct mlx4_ib_create_cq ucmd;
if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
err = -EFAULT;
goto err_cq;
}
buf_addr = (void *)(unsigned long)ucmd.buf_addr;
err = mlx4_ib_get_cq_umem(dev, udata, &cq->buf, &cq->umem,
ucmd.buf_addr, entries);
if (err)
goto err_cq;
err = mlx4_ib_db_map_user(udata, ucmd.db_addr, &cq->db);
if (err)
goto err_mtt;
uar = &context->uar;
cq->mcq.usage = MLX4_RES_USAGE_USER_VERBS;
} else {
err = mlx4_db_alloc(dev->dev, &cq->db, 1);
if (err)
goto err_cq;
cq->mcq.set_ci_db = cq->db.db;
cq->mcq.arm_db = cq->db.db + 1;
*cq->mcq.set_ci_db = 0;
*cq->mcq.arm_db = 0;
err = mlx4_ib_alloc_cq_buf(dev, &cq->buf, entries);
if (err)
goto err_db;
buf_addr = &cq->buf.buf;
uar = &dev->priv_uar;
cq->mcq.usage = MLX4_RES_USAGE_DRIVER;
}
if (dev->eq_table)
vector = dev->eq_table[vector % ibdev->num_comp_vectors];
err = mlx4_cq_alloc(dev->dev, entries, &cq->buf.mtt, uar, cq->db.dma,
&cq->mcq, vector, 0,
!!(cq->create_flags &
IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION),
buf_addr, !!udata);
if (err)
goto err_dbmap;
if (udata)
cq->mcq.tasklet_ctx.comp = mlx4_ib_cq_comp;
else
cq->mcq.comp = mlx4_ib_cq_comp;
cq->mcq.event = mlx4_ib_cq_event;
if (udata)
if (ib_copy_to_udata(udata, &cq->mcq.cqn, sizeof (__u32))) {
err = -EFAULT;
goto err_cq_free;
}
return &cq->ibcq;
err_cq_free:
mlx4_cq_free(dev->dev, &cq->mcq);
err_dbmap:
if (udata)
mlx4_ib_db_unmap_user(context, &cq->db);
err_mtt:
mlx4_mtt_cleanup(dev->dev, &cq->buf.mtt);
if (udata)
ib_umem_release(cq->umem);
else
mlx4_ib_free_cq_buf(dev, &cq->buf, cq->ibcq.cqe);
err_db:
if (!udata)
mlx4_db_free(dev->dev, &cq->db);
err_cq:
kfree(cq);
return ERR_PTR(err);
}
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;
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);
}
/* CQ allocation and modification test */
int cq_test(struct mlx4_dev *dev, char* log) {
struct mlx4_cq *cq;
struct mlx4_mtt *mtt;
struct mlx4_uar *uar;
struct mlx4_db *db;
int err;
int expected_rc = 0;
int collapsed = 0;
int timestamp_en = 0;
int npages = 1;
int page_shift = get_order(dev->caps.cqe_size) + PAGE_SHIFT;
int ret_val = FAIL;
int vector = 0;
int nent = 2 * MLX4_NUM_TUNNEL_BUFS;
u16 count = 88;
u16 period = 0;
u64 mtt_addr;
uar = malloc(sizeof *uar ,M_CQ_VAL, M_WAITOK );
VL_CHECK_MALLOC(uar, goto without_free, log);
mtt = malloc(sizeof *mtt ,M_CQ_VAL, M_WAITOK );
VL_CHECK_MALLOC(mtt, goto free_uar, log);
cq = malloc(sizeof *cq ,M_CQ_VAL, M_WAITOK );
VL_CHECK_MALLOC(cq, goto free_mtt, log);
db = malloc(sizeof *db ,M_CQ_VAL, M_WAITOK );
VL_CHECK_MALLOC(db, goto free_cq, log);
err = mlx4_mtt_init(dev, npages, page_shift, mtt);
VL_CHECK_RC(err, expected_rc, goto free_db , log, "failed to initialize MTT");
uprintf("MTT was initialized successfuly\n");
VL_CHECK_INT_VALUE(mtt->order, 0, goto cleanup_mtt, log, "mtt->order is wrong");
VL_CHECK_INT_VALUE(mtt->page_shift, 12, goto cleanup_mtt, log, "mtt->page_shift is wrong");
mtt_addr = mlx4_mtt_addr(dev, mtt);
uprintf("MTT address is: %lu\n", mtt_addr);
err = mlx4_uar_alloc(dev, uar);
VL_CHECK_RC(err, expected_rc, goto cleanup_mtt , log, "failed to allocate UAR");
uprintf("UAR was allocated successfuly\n");
err = mlx4_db_alloc(dev, db, 1);
VL_CHECK_RC(err, expected_rc, goto dealloc_uar , log, "failed to allocate DB");
uprintf("DB was allocated successfuly\n");
err = mlx4_cq_alloc(dev, nent, mtt, uar, db->dma, cq, vector, collapsed, timestamp_en);
VL_CHECK_RC(err, expected_rc, goto dealloc_db , log, "failed to allocate CQ");
uprintf("CQ allocated successfuly\n");
VL_CHECK_INT_VALUE(cq->cons_index, 0, goto dealloc_cq, log, "cq->cons_index is wrong");
VL_CHECK_INT_VALUE(cq->arm_sn, 1, goto dealloc_cq, log, "cq->arm_sn is wrong");
uprintf("cq->cqn = %d, cq->uar->pfn = %lu, cq->eqn = %d, cq->irq = %u\n", cq->cqn, cq->uar->pfn, cq->eqn, cq->irq );
VL_CHECK_UNSIGNED_INT_VALUE(cq->cons_index, (unsigned int)0, goto dealloc_cq, log, "cq->cons_index != 0");
VL_CHECK_INT_VALUE(cq->arm_sn, 1, goto dealloc_cq, log, "cq->arm_sn != 1");
err = mlx4_cq_modify(dev, cq, count, period);
VL_CHECK_RC(err, expected_rc, goto dealloc_cq , log, "failed to modify CQ");
uprintf("CQ was modifyed successfuly\n");
ret_val = SUCCESS;
dealloc_cq:
mlx4_cq_free(dev, cq);
uprintf("CQ was freed successfuly\n");
dealloc_db:
mlx4_db_free(dev, db);
uprintf( "DB free was successful\n");
dealloc_uar:
mlx4_uar_free(dev,uar);
uprintf("UAR free was successful\n");
cleanup_mtt:
mlx4_mtt_cleanup(dev, mtt);
uprintf( "mtt clean-up was successful\n");
free_db:
free(db, M_CQ_VAL);
free_cq:
free(cq, M_CQ_VAL);
free_mtt:
free(mtt, M_CQ_VAL);
free_uar:
free(uar, M_CQ_VAL);
without_free:
return ret_val;
}