/**
* rvt_create_cq - create a completion queue
* @ibdev: the device this completion queue is attached to
* @attr: creation attributes
* @context: unused by the QLogic_IB driver
* @udata: user data for libibverbs.so
*
* Called by ib_create_cq() in the generic verbs code.
*
* Return: pointer to the completion queue or negative errno values
* for failure.
*/
struct ib_cq *rvt_create_cq(struct ib_device *ibdev,
const struct ib_cq_init_attr *attr,
struct ib_ucontext *context,
struct ib_udata *udata)
{
struct rvt_dev_info *rdi = ib_to_rvt(ibdev);
struct rvt_cq *cq;
struct rvt_cq_wc *wc;
struct ib_cq *ret;
u32 sz;
unsigned int entries = attr->cqe;
if (attr->flags)
return ERR_PTR(-EINVAL);
if (entries < 1 || entries > rdi->dparms.props.max_cqe)
return ERR_PTR(-EINVAL);
/* Allocate the completion queue structure. */
cq = kzalloc(sizeof(*cq), GFP_KERNEL);
if (!cq)
return ERR_PTR(-ENOMEM);
/*
* Allocate the completion queue entries and head/tail pointers.
* This is allocated separately so that it can be resized and
* also mapped into user space.
* We need to use vmalloc() in order to support mmap and large
* numbers of entries.
*/
sz = sizeof(*wc);
if (udata && udata->outlen >= sizeof(__u64))
sz += sizeof(struct ib_uverbs_wc) * (entries + 1);
else
sz += sizeof(struct ib_wc) * (entries + 1);
wc = vmalloc_user(sz);
if (!wc) {
ret = ERR_PTR(-ENOMEM);
goto bail_cq;
}
/*
* Return the address of the WC as the offset to mmap.
* See rvt_mmap() for details.
*/
if (udata && udata->outlen >= sizeof(__u64)) {
int err;
cq->ip = rvt_create_mmap_info(rdi, sz, context, wc);
if (!cq->ip) {
ret = ERR_PTR(-ENOMEM);
goto bail_wc;
}
err = ib_copy_to_udata(udata, &cq->ip->offset,
sizeof(cq->ip->offset));
if (err) {
ret = ERR_PTR(err);
goto bail_ip;
}
}
spin_lock(&rdi->n_cqs_lock);
if (rdi->n_cqs_allocated == rdi->dparms.props.max_cq) {
spin_unlock(&rdi->n_cqs_lock);
ret = ERR_PTR(-ENOMEM);
goto bail_ip;
}
rdi->n_cqs_allocated++;
spin_unlock(&rdi->n_cqs_lock);
if (cq->ip) {
spin_lock_irq(&rdi->pending_lock);
list_add(&cq->ip->pending_mmaps, &rdi->pending_mmaps);
spin_unlock_irq(&rdi->pending_lock);
}
/*
* ib_create_cq() will initialize cq->ibcq except for cq->ibcq.cqe.
* The number of entries should be >= the number requested or return
* an error.
*/
cq->rdi = rdi;
cq->ibcq.cqe = entries;
cq->notify = RVT_CQ_NONE;
spin_lock_init(&cq->lock);
init_kthread_work(&cq->comptask, send_complete);
cq->queue = wc;
ret = &cq->ibcq;
goto done;
bail_ip:
kfree(cq->ip);
bail_wc:
vfree(wc);
bail_cq:
kfree(cq);
done:
return ret;
}
/**
* rvt_create_qp - create a queue pair for a device
* @ibpd: the protection domain who's device we create the queue pair for
* @init_attr: the attributes of the queue pair
* @udata: user data for libibverbs.so
*
* Queue pair creation is mostly an rvt issue. However, drivers have their own
* unique idea of what queue pair numbers mean. For instance there is a reserved
* range for PSM.
*
* Return: the queue pair on success, otherwise returns an errno.
*
* Called by the ib_create_qp() core verbs function.
*/
struct ib_qp *rvt_create_qp(struct ib_pd *ibpd,
struct ib_qp_init_attr *init_attr,
struct ib_udata *udata)
{
struct rvt_qp *qp;
int err;
struct rvt_swqe *swq = NULL;
size_t sz;
size_t sg_list_sz;
struct ib_qp *ret = ERR_PTR(-ENOMEM);
struct rvt_dev_info *rdi = ib_to_rvt(ibpd->device);
void *priv = NULL;
gfp_t gfp;
if (!rdi)
return ERR_PTR(-EINVAL);
if (init_attr->cap.max_send_sge > rdi->dparms.props.max_sge ||
init_attr->cap.max_send_wr > rdi->dparms.props.max_qp_wr ||
init_attr->create_flags & ~(IB_QP_CREATE_USE_GFP_NOIO))
return ERR_PTR(-EINVAL);
/* GFP_NOIO is applicable to RC QP's only */
if (init_attr->create_flags & IB_QP_CREATE_USE_GFP_NOIO &&
init_attr->qp_type != IB_QPT_RC)
return ERR_PTR(-EINVAL);
gfp = init_attr->create_flags & IB_QP_CREATE_USE_GFP_NOIO ?
GFP_NOIO : GFP_KERNEL;
/* Check receive queue parameters if no SRQ is specified. */
if (!init_attr->srq) {
if (init_attr->cap.max_recv_sge > rdi->dparms.props.max_sge ||
init_attr->cap.max_recv_wr > rdi->dparms.props.max_qp_wr)
return ERR_PTR(-EINVAL);
if (init_attr->cap.max_send_sge +
init_attr->cap.max_send_wr +
init_attr->cap.max_recv_sge +
init_attr->cap.max_recv_wr == 0)
return ERR_PTR(-EINVAL);
}
switch (init_attr->qp_type) {
case IB_QPT_SMI:
case IB_QPT_GSI:
if (init_attr->port_num == 0 ||
init_attr->port_num > ibpd->device->phys_port_cnt)
return ERR_PTR(-EINVAL);
case IB_QPT_UC:
case IB_QPT_RC:
case IB_QPT_UD:
sz = sizeof(struct rvt_sge) *
init_attr->cap.max_send_sge +
sizeof(struct rvt_swqe);
if (gfp == GFP_NOIO)
swq = __vmalloc(
(init_attr->cap.max_send_wr + 1) * sz,
gfp, PAGE_KERNEL);
else
swq = vmalloc_node(
(init_attr->cap.max_send_wr + 1) * sz,
rdi->dparms.node);
if (!swq)
return ERR_PTR(-ENOMEM);
sz = sizeof(*qp);
sg_list_sz = 0;
if (init_attr->srq) {
struct rvt_srq *srq = ibsrq_to_rvtsrq(init_attr->srq);
if (srq->rq.max_sge > 1)
sg_list_sz = sizeof(*qp->r_sg_list) *
(srq->rq.max_sge - 1);
} else if (init_attr->cap.max_recv_sge > 1)
sg_list_sz = sizeof(*qp->r_sg_list) *
(init_attr->cap.max_recv_sge - 1);
qp = kzalloc_node(sz + sg_list_sz, gfp, rdi->dparms.node);
if (!qp)
goto bail_swq;
RCU_INIT_POINTER(qp->next, NULL);
/*
* Driver needs to set up it's private QP structure and do any
* initialization that is needed.
*/
priv = rdi->driver_f.qp_priv_alloc(rdi, qp, gfp);
if (!priv)
goto bail_qp;
qp->priv = priv;
qp->timeout_jiffies =
usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /
1000UL);
if (init_attr->srq) {
sz = 0;
} else {
qp->r_rq.size = init_attr->cap.max_recv_wr + 1;
qp->r_rq.max_sge = init_attr->cap.max_recv_sge;
sz = (sizeof(struct ib_sge) * qp->r_rq.max_sge) +
sizeof(struct rvt_rwqe);
if (udata)
qp->r_rq.wq = vmalloc_user(
sizeof(struct rvt_rwq) +
qp->r_rq.size * sz);
else if (gfp == GFP_NOIO)
qp->r_rq.wq = __vmalloc(
sizeof(struct rvt_rwq) +
qp->r_rq.size * sz,
gfp, PAGE_KERNEL);
else
qp->r_rq.wq = vmalloc_node(
sizeof(struct rvt_rwq) +
qp->r_rq.size * sz,
rdi->dparms.node);
if (!qp->r_rq.wq)
goto bail_driver_priv;
}
/*
* ib_create_qp() will initialize qp->ibqp
* except for qp->ibqp.qp_num.
*/
spin_lock_init(&qp->r_lock);
spin_lock_init(&qp->s_hlock);
spin_lock_init(&qp->s_lock);
spin_lock_init(&qp->r_rq.lock);
atomic_set(&qp->refcount, 0);
init_waitqueue_head(&qp->wait);
init_timer(&qp->s_timer);
qp->s_timer.data = (unsigned long)qp;
INIT_LIST_HEAD(&qp->rspwait);
qp->state = IB_QPS_RESET;
qp->s_wq = swq;
qp->s_size = init_attr->cap.max_send_wr + 1;
qp->s_avail = init_attr->cap.max_send_wr;
qp->s_max_sge = init_attr->cap.max_send_sge;
if (init_attr->sq_sig_type == IB_SIGNAL_REQ_WR)
qp->s_flags = RVT_S_SIGNAL_REQ_WR;
err = alloc_qpn(rdi, &rdi->qp_dev->qpn_table,
init_attr->qp_type,
init_attr->port_num, gfp);
if (err < 0) {
ret = ERR_PTR(err);
goto bail_rq_wq;
}
qp->ibqp.qp_num = err;
qp->port_num = init_attr->port_num;
rvt_reset_qp(rdi, qp, init_attr->qp_type);
break;
default:
/* Don't support raw QPs */
return ERR_PTR(-EINVAL);
}
init_attr->cap.max_inline_data = 0;
/*
* Return the address of the RWQ as the offset to mmap.
* See rvt_mmap() for details.
*/
if (udata && udata->outlen >= sizeof(__u64)) {
if (!qp->r_rq.wq) {
__u64 offset = 0;
err = ib_copy_to_udata(udata, &offset,
sizeof(offset));
if (err) {
ret = ERR_PTR(err);
goto bail_qpn;
}
} else {
u32 s = sizeof(struct rvt_rwq) + qp->r_rq.size * sz;
qp->ip = rvt_create_mmap_info(rdi, s,
ibpd->uobject->context,
qp->r_rq.wq);
if (!qp->ip) {
ret = ERR_PTR(-ENOMEM);
goto bail_qpn;
}
err = ib_copy_to_udata(udata, &qp->ip->offset,
sizeof(qp->ip->offset));
if (err) {
ret = ERR_PTR(err);
goto bail_ip;
}
}
qp->pid = current->pid;
}
spin_lock(&rdi->n_qps_lock);
if (rdi->n_qps_allocated == rdi->dparms.props.max_qp) {
spin_unlock(&rdi->n_qps_lock);
ret = ERR_PTR(-ENOMEM);
goto bail_ip;
}
rdi->n_qps_allocated++;
/*
* Maintain a busy_jiffies variable that will be added to the timeout
* period in mod_retry_timer and add_retry_timer. This busy jiffies
* is scaled by the number of rc qps created for the device to reduce
* the number of timeouts occurring when there is a large number of
* qps. busy_jiffies is incremented every rc qp scaling interval.
* The scaling interval is selected based on extensive performance
* evaluation of targeted workloads.
*/
if (init_attr->qp_type == IB_QPT_RC) {
rdi->n_rc_qps++;
rdi->busy_jiffies = rdi->n_rc_qps / RC_QP_SCALING_INTERVAL;
}
spin_unlock(&rdi->n_qps_lock);
if (qp->ip) {
spin_lock_irq(&rdi->pending_lock);
list_add(&qp->ip->pending_mmaps, &rdi->pending_mmaps);
spin_unlock_irq(&rdi->pending_lock);
}
ret = &qp->ibqp;
/*
* We have our QP and its good, now keep track of what types of opcodes
* can be processed on this QP. We do this by keeping track of what the
* 3 high order bits of the opcode are.
*/
switch (init_attr->qp_type) {
case IB_QPT_SMI:
case IB_QPT_GSI:
case IB_QPT_UD:
qp->allowed_ops = IB_OPCODE_UD_SEND_ONLY & RVT_OPCODE_QP_MASK;
break;
case IB_QPT_RC:
qp->allowed_ops = IB_OPCODE_RC_SEND_ONLY & RVT_OPCODE_QP_MASK;
break;
case IB_QPT_UC:
qp->allowed_ops = IB_OPCODE_UC_SEND_ONLY & RVT_OPCODE_QP_MASK;
break;
default:
ret = ERR_PTR(-EINVAL);
goto bail_ip;
}
return ret;
bail_ip:
kref_put(&qp->ip->ref, rvt_release_mmap_info);
bail_qpn:
free_qpn(&rdi->qp_dev->qpn_table, qp->ibqp.qp_num);
bail_rq_wq:
vfree(qp->r_rq.wq);
bail_driver_priv:
rdi->driver_f.qp_priv_free(rdi, qp);
bail_qp:
kfree(qp);
bail_swq:
vfree(swq);
return ret;
}
/**
* rvt_create_srq - create a shared receive queue
* @ibpd: the protection domain of the SRQ to create
* @srq_init_attr: the attributes of the SRQ
* @udata: data from libibverbs when creating a user SRQ
*
* Return: Allocated srq object
*/
struct ib_srq *rvt_create_srq(struct ib_pd *ibpd,
struct ib_srq_init_attr *srq_init_attr,
struct ib_udata *udata)
{
struct rvt_dev_info *dev = ib_to_rvt(ibpd->device);
struct rvt_ucontext *ucontext = rdma_udata_to_drv_context(
udata, struct rvt_ucontext, ibucontext);
struct rvt_srq *srq;
u32 sz;
struct ib_srq *ret;
if (srq_init_attr->srq_type != IB_SRQT_BASIC)
return ERR_PTR(-EOPNOTSUPP);
if (srq_init_attr->attr.max_sge == 0 ||
srq_init_attr->attr.max_sge > dev->dparms.props.max_srq_sge ||
srq_init_attr->attr.max_wr == 0 ||
srq_init_attr->attr.max_wr > dev->dparms.props.max_srq_wr)
return ERR_PTR(-EINVAL);
srq = kzalloc_node(sizeof(*srq), GFP_KERNEL, dev->dparms.node);
if (!srq)
return ERR_PTR(-ENOMEM);
/*
* Need to use vmalloc() if we want to support large #s of entries.
*/
srq->rq.size = srq_init_attr->attr.max_wr + 1;
srq->rq.max_sge = srq_init_attr->attr.max_sge;
sz = sizeof(struct ib_sge) * srq->rq.max_sge +
sizeof(struct rvt_rwqe);
srq->rq.wq = udata ?
vmalloc_user(sizeof(struct rvt_rwq) + srq->rq.size * sz) :
vzalloc_node(sizeof(struct rvt_rwq) + srq->rq.size * sz,
dev->dparms.node);
if (!srq->rq.wq) {
ret = ERR_PTR(-ENOMEM);
goto bail_srq;
}
/*
* Return the address of the RWQ as the offset to mmap.
* See rvt_mmap() for details.
*/
if (udata && udata->outlen >= sizeof(__u64)) {
int err;
u32 s = sizeof(struct rvt_rwq) + srq->rq.size * sz;
srq->ip =
rvt_create_mmap_info(dev, s, &ucontext->ibucontext,
srq->rq.wq);
if (!srq->ip) {
ret = ERR_PTR(-ENOMEM);
goto bail_wq;
}
err = ib_copy_to_udata(udata, &srq->ip->offset,
sizeof(srq->ip->offset));
if (err) {
ret = ERR_PTR(err);
goto bail_ip;
}
}
/*
* ib_create_srq() will initialize srq->ibsrq.
*/
spin_lock_init(&srq->rq.lock);
srq->limit = srq_init_attr->attr.srq_limit;
spin_lock(&dev->n_srqs_lock);
if (dev->n_srqs_allocated == dev->dparms.props.max_srq) {
spin_unlock(&dev->n_srqs_lock);
ret = ERR_PTR(-ENOMEM);
goto bail_ip;
}
dev->n_srqs_allocated++;
spin_unlock(&dev->n_srqs_lock);
if (srq->ip) {
spin_lock_irq(&dev->pending_lock);
list_add(&srq->ip->pending_mmaps, &dev->pending_mmaps);
spin_unlock_irq(&dev->pending_lock);
}
return &srq->ibsrq;
bail_ip:
kfree(srq->ip);
bail_wq:
vfree(srq->rq.wq);
bail_srq:
kfree(srq);
return ret;
}
