/**
* ipath_destroy_srq - destroy a shared receive queue
* @ibsrq: the SRQ to destroy
*/
int ipath_destroy_srq(struct ib_srq *ibsrq)
{
struct ipath_srq *srq = to_isrq(ibsrq);
vfree(srq->rq.wq);
kfree(srq);
return 0;
}
int qib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr)
{
struct qib_srq *srq = to_isrq(ibsrq);
attr->max_wr = srq->rq.size - 1;
attr->max_sge = srq->rq.max_sge;
attr->srq_limit = srq->limit;
return 0;
}
/**
* ipath_destroy_srq - destroy a shared receive queue
* @ibsrq: the SRQ to destroy
*/
int ipath_destroy_srq(struct ib_srq *ibsrq)
{
struct ipath_srq *srq = to_isrq(ibsrq);
struct ipath_ibdev *dev = to_idev(ibsrq->device);
dev->n_srqs_allocated--;
vfree(srq->rq.wq);
kfree(srq);
return 0;
}
/**
* qib_destroy_srq - destroy a shared receive queue
* @ibsrq: the SRQ to destroy
*/
int qib_destroy_srq(struct ib_srq *ibsrq)
{
struct qib_srq *srq = to_isrq(ibsrq);
struct qib_ibdev *dev = to_idev(ibsrq->device);
spin_lock(&dev->n_srqs_lock);
dev->n_srqs_allocated--;
spin_unlock(&dev->n_srqs_lock);
if (srq->ip)
kref_put(&srq->ip->ref, qib_release_mmap_info);
else
vfree(srq->rq.wq);
kfree(srq);
return 0;
}
/**
* qib_post_srq_receive - post a receive on a shared receive queue
* @ibsrq: the SRQ to post the receive on
* @wr: the list of work requests to post
* @bad_wr: A pointer to the first WR to cause a problem is put here
*
* This may be called from interrupt context.
*/
int qib_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
struct ib_recv_wr **bad_wr)
{
struct qib_srq *srq = to_isrq(ibsrq);
struct qib_rwq *wq;
unsigned long flags;
int ret;
for (; wr; wr = wr->next) {
struct qib_rwqe *wqe;
u32 next;
int i;
if ((unsigned) wr->num_sge > srq->rq.max_sge) {
*bad_wr = wr;
ret = -EINVAL;
goto bail;
}
spin_lock_irqsave(&srq->rq.lock, flags);
wq = srq->rq.wq;
next = wq->head + 1;
if (next >= srq->rq.size)
next = 0;
if (next == wq->tail) {
spin_unlock_irqrestore(&srq->rq.lock, flags);
*bad_wr = wr;
ret = -ENOMEM;
goto bail;
}
wqe = get_rwqe_ptr(&srq->rq, wq->head);
wqe->wr_id = wr->wr_id;
wqe->num_sge = wr->num_sge;
for (i = 0; i < wr->num_sge; i++)
wqe->sg_list[i] = wr->sg_list[i];
/* Make sure queue entry is written before the head index. */
smp_wmb();
wq->head = next;
spin_unlock_irqrestore(&srq->rq.lock, flags);
}
ret = 0;
bail:
return ret;
}
/**
* qib_modify_srq - modify a shared receive queue
* @ibsrq: the SRQ to modify
* @attr: the new attributes of the SRQ
* @attr_mask: indicates which attributes to modify
* @udata: user data for libibverbs.so
*/
int qib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
enum ib_srq_attr_mask attr_mask,
struct ib_udata *udata)
{
struct qib_srq *srq = to_isrq(ibsrq);
struct qib_rwq *wq;
int ret = 0;
if (attr_mask & IB_SRQ_MAX_WR) {
struct qib_rwq *owq;
struct qib_rwqe *p;
u32 sz, size, n, head, tail;
/* Check that the requested sizes are below the limits. */
if ((attr->max_wr > ib_qib_max_srq_wrs) ||
((attr_mask & IB_SRQ_LIMIT) ?
attr->srq_limit : srq->limit) > attr->max_wr) {
ret = -EINVAL;
goto bail;
}
sz = sizeof(struct qib_rwqe) +
srq->rq.max_sge * sizeof(struct ib_sge);
size = attr->max_wr + 1;
wq = vmalloc_user(sizeof(struct qib_rwq) + size * sz);
if (!wq) {
ret = -ENOMEM;
goto bail;
}
/* Check that we can write the offset to mmap. */
if (udata && udata->inlen >= sizeof(__u64)) {
__u64 offset_addr;
__u64 offset = 0;
ret = ib_copy_from_udata(&offset_addr, udata,
sizeof(offset_addr));
if (ret)
goto bail_free;
udata->outbuf =
(void __user *) (unsigned long) offset_addr;
ret = ib_copy_to_udata(udata, &offset,
sizeof(offset));
if (ret)
goto bail_free;
}
spin_lock_irq(&srq->rq.lock);
/*
* validate head and tail pointer values and compute
* the number of remaining WQEs.
*/
owq = srq->rq.wq;
head = owq->head;
tail = owq->tail;
if (head >= srq->rq.size || tail >= srq->rq.size) {
ret = -EINVAL;
goto bail_unlock;
}
n = head;
if (n < tail)
n += srq->rq.size - tail;
else
n -= tail;
if (size <= n) {
ret = -EINVAL;
goto bail_unlock;
}
n = 0;
p = wq->wq;
while (tail != head) {
struct qib_rwqe *wqe;
int i;
wqe = get_rwqe_ptr(&srq->rq, tail);
p->wr_id = wqe->wr_id;
p->num_sge = wqe->num_sge;
for (i = 0; i < wqe->num_sge; i++)
p->sg_list[i] = wqe->sg_list[i];
n++;
p = (struct qib_rwqe *)((char *) p + sz);
if (++tail >= srq->rq.size)
tail = 0;
}
srq->rq.wq = wq;
srq->rq.size = size;
wq->head = n;
wq->tail = 0;
if (attr_mask & IB_SRQ_LIMIT)
srq->limit = attr->srq_limit;
spin_unlock_irq(&srq->rq.lock);
vfree(owq);
if (srq->ip) {
struct qib_mmap_info *ip = srq->ip;
struct qib_ibdev *dev = to_idev(srq->ibsrq.device);
u32 s = sizeof(struct qib_rwq) + size * sz;
qib_update_mmap_info(dev, ip, s, wq);
/*
* Return the offset to mmap.
* See qib_mmap() for details.
*/
if (udata && udata->inlen >= sizeof(__u64)) {
ret = ib_copy_to_udata(udata, &ip->offset,
sizeof(ip->offset));
if (ret)
goto bail;
}
/*
* Put user mapping info onto the pending list
* unless it already is on the list.
*/
spin_lock_irq(&dev->pending_lock);
if (list_empty(&ip->pending_mmaps))
list_add(&ip->pending_mmaps,
&dev->pending_mmaps);
spin_unlock_irq(&dev->pending_lock);
}
} else if (attr_mask & IB_SRQ_LIMIT) {
spin_lock_irq(&srq->rq.lock);
if (attr->srq_limit >= srq->rq.size)
ret = -EINVAL;
else
srq->limit = attr->srq_limit;
spin_unlock_irq(&srq->rq.lock);
}
goto bail;
bail_unlock:
spin_unlock_irq(&srq->rq.lock);
bail_free:
vfree(wq);
bail:
return ret;
}
struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
struct ib_qp_init_attr *init_attr,
struct ib_udata *udata)
{
struct ipath_qp *qp;
int err;
struct ipath_swqe *swq = NULL;
struct ipath_ibdev *dev;
size_t sz;
size_t sg_list_sz;
struct ib_qp *ret;
if (init_attr->create_flags) {
ret = ERR_PTR(-EINVAL);
goto bail;
}
if (init_attr->cap.max_send_sge > ib_ipath_max_sges ||
init_attr->cap.max_send_wr > ib_ipath_max_qp_wrs) {
ret = ERR_PTR(-EINVAL);
goto bail;
}
/* */
if (!init_attr->srq) {
if (init_attr->cap.max_recv_sge > ib_ipath_max_sges ||
init_attr->cap.max_recv_wr > ib_ipath_max_qp_wrs) {
ret = ERR_PTR(-EINVAL);
goto bail;
}
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) {
ret = ERR_PTR(-EINVAL);
goto bail;
}
}
switch (init_attr->qp_type) {
case IB_QPT_UC:
case IB_QPT_RC:
case IB_QPT_UD:
case IB_QPT_SMI:
case IB_QPT_GSI:
sz = sizeof(struct ipath_sge) *
init_attr->cap.max_send_sge +
sizeof(struct ipath_swqe);
swq = vmalloc((init_attr->cap.max_send_wr + 1) * sz);
if (swq == NULL) {
ret = ERR_PTR(-ENOMEM);
goto bail;
}
sz = sizeof(*qp);
sg_list_sz = 0;
if (init_attr->srq) {
struct ipath_srq *srq = to_isrq(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 = kmalloc(sz + sg_list_sz, GFP_KERNEL);
if (!qp) {
ret = ERR_PTR(-ENOMEM);
goto bail_swq;
}
if (sg_list_sz && (init_attr->qp_type == IB_QPT_UD ||
init_attr->qp_type == IB_QPT_SMI ||
init_attr->qp_type == IB_QPT_GSI)) {
qp->r_ud_sg_list = kmalloc(sg_list_sz, GFP_KERNEL);
if (!qp->r_ud_sg_list) {
ret = ERR_PTR(-ENOMEM);
goto bail_qp;
}
} else
qp->r_ud_sg_list = NULL;
if (init_attr->srq) {
sz = 0;
qp->r_rq.size = 0;
qp->r_rq.max_sge = 0;
qp->r_rq.wq = NULL;
init_attr->cap.max_recv_wr = 0;
init_attr->cap.max_recv_sge = 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 ipath_rwqe);
qp->r_rq.wq = vmalloc_user(sizeof(struct ipath_rwq) +
qp->r_rq.size * sz);
if (!qp->r_rq.wq) {
ret = ERR_PTR(-ENOMEM);
goto bail_sg_list;
}
}
/*
*/
spin_lock_init(&qp->s_lock);
spin_lock_init(&qp->r_rq.lock);
atomic_set(&qp->refcount, 0);
init_waitqueue_head(&qp->wait);
init_waitqueue_head(&qp->wait_dma);
tasklet_init(&qp->s_task, ipath_do_send, (unsigned long)qp);
INIT_LIST_HEAD(&qp->piowait);
INIT_LIST_HEAD(&qp->timerwait);
qp->state = IB_QPS_RESET;
qp->s_wq = swq;
qp->s_size = init_attr->cap.max_send_wr + 1;
qp->s_max_sge = init_attr->cap.max_send_sge;
if (init_attr->sq_sig_type == IB_SIGNAL_REQ_WR)
qp->s_flags = IPATH_S_SIGNAL_REQ_WR;
else
qp->s_flags = 0;
dev = to_idev(ibpd->device);
err = ipath_alloc_qpn(&dev->qp_table, qp,
init_attr->qp_type);
if (err) {
ret = ERR_PTR(err);
vfree(qp->r_rq.wq);
goto bail_sg_list;
}
qp->ip = NULL;
qp->s_tx = NULL;
ipath_reset_qp(qp, init_attr->qp_type);
break;
default:
/* */
ret = ERR_PTR(-ENOSYS);
goto bail;
}
init_attr->cap.max_inline_data = 0;
/*
*/
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_ip;
}
} else {
u32 s = sizeof(struct ipath_rwq) +
qp->r_rq.size * sz;
qp->ip =
ipath_create_mmap_info(dev, s,
ibpd->uobject->context,
qp->r_rq.wq);
if (!qp->ip) {
ret = ERR_PTR(-ENOMEM);
goto bail_ip;
}
err = ib_copy_to_udata(udata, &(qp->ip->offset),
sizeof(qp->ip->offset));
if (err) {
ret = ERR_PTR(err);
goto bail_ip;
}
}
}
spin_lock(&dev->n_qps_lock);
if (dev->n_qps_allocated == ib_ipath_max_qps) {
spin_unlock(&dev->n_qps_lock);
ret = ERR_PTR(-ENOMEM);
goto bail_ip;
}
dev->n_qps_allocated++;
spin_unlock(&dev->n_qps_lock);
if (qp->ip) {
spin_lock_irq(&dev->pending_lock);
list_add(&qp->ip->pending_mmaps, &dev->pending_mmaps);
spin_unlock_irq(&dev->pending_lock);
}
ret = &qp->ibqp;
goto bail;
bail_ip:
if (qp->ip)
kref_put(&qp->ip->ref, ipath_release_mmap_info);
else
vfree(qp->r_rq.wq);
ipath_free_qp(&dev->qp_table, qp);
free_qpn(&dev->qp_table, qp->ibqp.qp_num);
bail_sg_list:
kfree(qp->r_ud_sg_list);
bail_qp:
kfree(qp);
bail_swq:
vfree(swq);
bail:
return ret;
}
/**
* qib_get_rwqe - copy the next RWQE into the QP's RWQE
* @qp: the QP
* @wr_id_only: update qp->r_wr_id only, not qp->r_sge
*
* Return -1 if there is a local error, 0 if no RWQE is available,
* otherwise return 1.
*
* Can be called from interrupt level.
*/
int qib_get_rwqe(struct qib_qp *qp, int wr_id_only)
{
unsigned long flags;
struct qib_rq *rq;
struct qib_rwq *wq;
struct qib_srq *srq;
struct qib_rwqe *wqe;
void (*handler)(struct ib_event *, void *);
u32 tail;
int ret;
if (qp->ibqp.srq) {
srq = to_isrq(qp->ibqp.srq);
handler = srq->ibsrq.event_handler;
rq = &srq->rq;
} else {
srq = NULL;
handler = NULL;
rq = &qp->r_rq;
}
spin_lock_irqsave(&rq->lock, flags);
if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)) {
ret = 0;
goto unlock;
}
wq = rq->wq;
tail = wq->tail;
/* Validate tail before using it since it is user writable. */
if (tail >= rq->size)
tail = 0;
if (unlikely(tail == wq->head)) {
ret = 0;
goto unlock;
}
/* Make sure entry is read after head index is read. */
smp_rmb();
wqe = get_rwqe_ptr(rq, tail);
/*
* Even though we update the tail index in memory, the verbs
* consumer is not supposed to post more entries until a
* completion is generated.
*/
if (++tail >= rq->size)
tail = 0;
wq->tail = tail;
if (!wr_id_only && !qib_init_sge(qp, wqe)) {
ret = -1;
goto unlock;
}
qp->r_wr_id = wqe->wr_id;
ret = 1;
set_bit(QIB_R_WRID_VALID, &qp->r_aflags);
if (handler) {
u32 n;
/*
* Validate head pointer value and compute
* the number of remaining WQEs.
*/
n = wq->head;
if (n >= rq->size)
n = 0;
if (n < tail)
n += rq->size - tail;
else
n -= tail;
if (n < srq->limit) {
struct ib_event ev;
srq->limit = 0;
spin_unlock_irqrestore(&rq->lock, flags);
ev.device = qp->ibqp.device;
ev.element.srq = qp->ibqp.srq;
ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
handler(&ev, srq->ibsrq.srq_context);
goto bail;
}
}
unlock:
spin_unlock_irqrestore(&rq->lock, flags);
bail:
return ret;
}
/**
* ipath_get_rwqe - copy the next RWQE into the QP's RWQE
* @qp: the QP
* @wr_id_only: update wr_id only, not SGEs
*
* Return 0 if no RWQE is available, otherwise return 1.
*
* Can be called from interrupt level.
*/
int ipath_get_rwqe(struct ipath_qp *qp, int wr_id_only)
{
unsigned long flags;
struct ipath_rq *rq;
struct ipath_rwq *wq;
struct ipath_srq *srq;
struct ipath_rwqe *wqe;
void (*handler)(struct ib_event *, void *);
u32 tail;
int ret;
qp->r_sge.sg_list = qp->r_sg_list;
if (qp->ibqp.srq) {
srq = to_isrq(qp->ibqp.srq);
handler = srq->ibsrq.event_handler;
rq = &srq->rq;
} else {
srq = NULL;
handler = NULL;
rq = &qp->r_rq;
}
spin_lock_irqsave(&rq->lock, flags);
if (!(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_RECV_OK)) {
ret = 0;
goto unlock;
}
wq = rq->wq;
tail = wq->tail;
/* Validate tail before using it since it is user writable. */
if (tail >= rq->size)
tail = 0;
do {
if (unlikely(tail == wq->head)) {
ret = 0;
goto unlock;
}
/* Make sure entry is read after head index is read. */
smp_rmb();
wqe = get_rwqe_ptr(rq, tail);
if (++tail >= rq->size)
tail = 0;
} while (!wr_id_only && !ipath_init_sge(qp, wqe, &qp->r_len,
&qp->r_sge));
qp->r_wr_id = wqe->wr_id;
wq->tail = tail;
ret = 1;
set_bit(IPATH_R_WRID_VALID, &qp->r_aflags);
if (handler) {
u32 n;
/*
* validate head pointer value and compute
* the number of remaining WQEs.
*/
n = wq->head;
if (n >= rq->size)
n = 0;
if (n < tail)
n += rq->size - tail;
else
n -= tail;
if (n < srq->limit) {
struct ib_event ev;
srq->limit = 0;
spin_unlock_irqrestore(&rq->lock, flags);
ev.device = qp->ibqp.device;
ev.element.srq = qp->ibqp.srq;
ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
handler(&ev, srq->ibsrq.srq_context);
goto bail;
}
}
unlock:
spin_unlock_irqrestore(&rq->lock, flags);
bail:
return ret;
}
/**
* ipath_post_srq_receive - post a receive on a shared receive queue
* @ibsrq: the SRQ to post the receive on
* @wr: the list of work requests to post
* @bad_wr: the first WR to cause a problem is put here
*
* This may be called from interrupt context.
*/
int ipath_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
struct ib_recv_wr **bad_wr)
{
struct ipath_srq *srq = to_isrq(ibsrq);
struct ipath_ibdev *dev = to_idev(ibsrq->device);
unsigned long flags;
int ret;
for (; wr; wr = wr->next) {
struct ipath_rwqe *wqe;
u32 next;
int i, j;
if (wr->num_sge > srq->rq.max_sge) {
*bad_wr = wr;
ret = -ENOMEM;
goto bail;
}
spin_lock_irqsave(&srq->rq.lock, flags);
next = srq->rq.head + 1;
if (next >= srq->rq.size)
next = 0;
if (next == srq->rq.tail) {
spin_unlock_irqrestore(&srq->rq.lock, flags);
*bad_wr = wr;
ret = -ENOMEM;
goto bail;
}
wqe = get_rwqe_ptr(&srq->rq, srq->rq.head);
wqe->wr_id = wr->wr_id;
wqe->sg_list[0].mr = NULL;
wqe->sg_list[0].vaddr = NULL;
wqe->sg_list[0].length = 0;
wqe->sg_list[0].sge_length = 0;
wqe->length = 0;
for (i = 0, j = 0; i < wr->num_sge; i++) {
/* Check LKEY */
if (to_ipd(srq->ibsrq.pd)->user &&
wr->sg_list[i].lkey == 0) {
spin_unlock_irqrestore(&srq->rq.lock,
flags);
*bad_wr = wr;
ret = -EINVAL;
goto bail;
}
if (wr->sg_list[i].length == 0)
continue;
if (!ipath_lkey_ok(&dev->lk_table,
&wqe->sg_list[j],
&wr->sg_list[i],
IB_ACCESS_LOCAL_WRITE)) {
spin_unlock_irqrestore(&srq->rq.lock,
flags);
*bad_wr = wr;
ret = -EINVAL;
goto bail;
}
wqe->length += wr->sg_list[i].length;
j++;
}
wqe->num_sge = j;
srq->rq.head = next;
spin_unlock_irqrestore(&srq->rq.lock, flags);
}
ret = 0;
bail:
return ret;
}
/**
* ipath_modify_srq - modify a shared receive queue
* @ibsrq: the SRQ to modify
* @attr: the new attributes of the SRQ
* @attr_mask: indicates which attributes to modify
*/
int ipath_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
enum ib_srq_attr_mask attr_mask)
{
struct ipath_srq *srq = to_isrq(ibsrq);
unsigned long flags;
int ret;
if (attr_mask & IB_SRQ_MAX_WR)
if ((attr->max_wr > ib_ipath_max_srq_wrs) ||
(attr->max_sge > srq->rq.max_sge)) {
ret = -EINVAL;
goto bail;
}
if (attr_mask & IB_SRQ_LIMIT)
if (attr->srq_limit >= srq->rq.size) {
ret = -EINVAL;
goto bail;
}
if (attr_mask & IB_SRQ_MAX_WR) {
struct ipath_rwqe *wq, *p;
u32 sz, size, n;
sz = sizeof(struct ipath_rwqe) +
attr->max_sge * sizeof(struct ipath_sge);
size = attr->max_wr + 1;
wq = vmalloc(size * sz);
if (!wq) {
ret = -ENOMEM;
goto bail;
}
spin_lock_irqsave(&srq->rq.lock, flags);
if (srq->rq.head < srq->rq.tail)
n = srq->rq.size + srq->rq.head - srq->rq.tail;
else
n = srq->rq.head - srq->rq.tail;
if (size <= n || size <= srq->limit) {
spin_unlock_irqrestore(&srq->rq.lock, flags);
vfree(wq);
ret = -EINVAL;
goto bail;
}
n = 0;
p = wq;
while (srq->rq.tail != srq->rq.head) {
struct ipath_rwqe *wqe;
int i;
wqe = get_rwqe_ptr(&srq->rq, srq->rq.tail);
p->wr_id = wqe->wr_id;
p->length = wqe->length;
p->num_sge = wqe->num_sge;
for (i = 0; i < wqe->num_sge; i++)
p->sg_list[i] = wqe->sg_list[i];
n++;
p = (struct ipath_rwqe *)((char *) p + sz);
if (++srq->rq.tail >= srq->rq.size)
srq->rq.tail = 0;
}
vfree(srq->rq.wq);
srq->rq.wq = wq;
srq->rq.size = size;
srq->rq.head = n;
srq->rq.tail = 0;
srq->rq.max_sge = attr->max_sge;
spin_unlock_irqrestore(&srq->rq.lock, flags);
}
if (attr_mask & IB_SRQ_LIMIT) {
spin_lock_irqsave(&srq->rq.lock, flags);
srq->limit = attr->srq_limit;
spin_unlock_irqrestore(&srq->rq.lock, flags);
}
ret = 0;
bail:
return ret;
}
/**
* ipath_get_rwqe - copy the next RWQE into the QP's RWQE
* @qp: the QP
* @wr_id_only: update wr_id only, not SGEs
*
* Return 0 if no RWQE is available, otherwise return 1.
*
* Called at interrupt level with the QP r_rq.lock held.
*/
int ipath_get_rwqe(struct ipath_qp *qp, int wr_id_only)
{
struct ipath_rq *rq;
struct ipath_srq *srq;
struct ipath_rwqe *wqe;
int ret;
if (!qp->ibqp.srq) {
rq = &qp->r_rq;
if (unlikely(rq->tail == rq->head)) {
ret = 0;
goto bail;
}
wqe = get_rwqe_ptr(rq, rq->tail);
qp->r_wr_id = wqe->wr_id;
if (!wr_id_only) {
qp->r_sge.sge = wqe->sg_list[0];
qp->r_sge.sg_list = wqe->sg_list + 1;
qp->r_sge.num_sge = wqe->num_sge;
qp->r_len = wqe->length;
}
if (++rq->tail >= rq->size)
rq->tail = 0;
ret = 1;
goto bail;
}
srq = to_isrq(qp->ibqp.srq);
rq = &srq->rq;
spin_lock(&rq->lock);
if (unlikely(rq->tail == rq->head)) {
spin_unlock(&rq->lock);
ret = 0;
goto bail;
}
wqe = get_rwqe_ptr(rq, rq->tail);
qp->r_wr_id = wqe->wr_id;
if (!wr_id_only) {
qp->r_sge.sge = wqe->sg_list[0];
qp->r_sge.sg_list = wqe->sg_list + 1;
qp->r_sge.num_sge = wqe->num_sge;
qp->r_len = wqe->length;
}
if (++rq->tail >= rq->size)
rq->tail = 0;
if (srq->ibsrq.event_handler) {
struct ib_event ev;
u32 n;
if (rq->head < rq->tail)
n = rq->size + rq->head - rq->tail;
else
n = rq->head - rq->tail;
if (n < srq->limit) {
srq->limit = 0;
spin_unlock(&rq->lock);
ev.device = qp->ibqp.device;
ev.element.srq = qp->ibqp.srq;
ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
srq->ibsrq.event_handler(&ev,
srq->ibsrq.srq_context);
} else
spin_unlock(&rq->lock);
} else
spin_unlock(&rq->lock);
ret = 1;
bail:
return ret;
}
int qib_get_rwqe(struct qib_qp *qp, int wr_id_only)
{
unsigned long flags;
struct qib_rq *rq;
struct qib_rwq *wq;
struct qib_srq *srq;
struct qib_rwqe *wqe;
void (*handler)(struct ib_event *, void *);
u32 tail;
int ret;
if (qp->ibqp.srq) {
srq = to_isrq(qp->ibqp.srq);
handler = srq->ibsrq.event_handler;
rq = &srq->rq;
} else {
srq = NULL;
handler = NULL;
rq = &qp->r_rq;
}
spin_lock_irqsave(&rq->lock, flags);
if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)) {
ret = 0;
goto unlock;
}
wq = rq->wq;
tail = wq->tail;
if (tail >= rq->size)
tail = 0;
if (unlikely(tail == wq->head)) {
ret = 0;
goto unlock;
}
smp_rmb();
wqe = get_rwqe_ptr(rq, tail);
if (++tail >= rq->size)
tail = 0;
wq->tail = tail;
if (!wr_id_only && !qib_init_sge(qp, wqe)) {
ret = -1;
goto unlock;
}
qp->r_wr_id = wqe->wr_id;
ret = 1;
set_bit(QIB_R_WRID_VALID, &qp->r_aflags);
if (handler) {
u32 n;
n = wq->head;
if (n >= rq->size)
n = 0;
if (n < tail)
n += rq->size - tail;
else
n -= tail;
if (n < srq->limit) {
struct ib_event ev;
srq->limit = 0;
spin_unlock_irqrestore(&rq->lock, flags);
ev.device = qp->ibqp.device;
ev.element.srq = qp->ibqp.srq;
ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
handler(&ev, srq->ibsrq.srq_context);
goto bail;
}
}
unlock:
spin_unlock_irqrestore(&rq->lock, flags);
bail:
return ret;
}
/**
* ipath_get_rwqe - copy the next RWQE into the QP's RWQE
* @qp: the QP
* @wr_id_only: update wr_id only, not SGEs
*
* Return 0 if no RWQE is available, otherwise return 1.
*
* Can be called from interrupt level.
*/
int ipath_get_rwqe(struct ipath_qp *qp, int wr_id_only)
{
unsigned long flags;
struct ipath_rq *rq;
struct ipath_rwq *wq;
struct ipath_srq *srq;
struct ipath_rwqe *wqe;
void (*handler)(struct ib_event *, void *);
u32 tail;
int ret;
if (qp->ibqp.srq) {
srq = to_isrq(qp->ibqp.srq);
handler = srq->ibsrq.event_handler;
rq = &srq->rq;
} else {
srq = NULL;
handler = NULL;
rq = &qp->r_rq;
}
spin_lock_irqsave(&rq->lock, flags);
wq = rq->wq;
tail = wq->tail;
/* Validate tail before using it since it is user writable. */
if (tail >= rq->size)
tail = 0;
do {
if (unlikely(tail == wq->head)) {
spin_unlock_irqrestore(&rq->lock, flags);
ret = 0;
goto bail;
}
wqe = get_rwqe_ptr(rq, tail);
if (++tail >= rq->size)
tail = 0;
} while (!wr_id_only && !init_sge(qp, wqe));
qp->r_wr_id = wqe->wr_id;
wq->tail = tail;
ret = 1;
qp->r_wrid_valid = 1;
if (handler) {
u32 n;
/*
* validate head pointer value and compute
* the number of remaining WQEs.
*/
n = wq->head;
if (n >= rq->size)
n = 0;
if (n < tail)
n += rq->size - tail;
else
n -= tail;
if (n < srq->limit) {
struct ib_event ev;
srq->limit = 0;
spin_unlock_irqrestore(&rq->lock, flags);
ev.device = qp->ibqp.device;
ev.element.srq = qp->ibqp.srq;
ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
handler(&ev, srq->ibsrq.srq_context);
goto bail;
}
}
spin_unlock_irqrestore(&rq->lock, flags);
bail:
return ret;
}
/**
* hfi1_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
*
* Returns the queue pair on success, otherwise returns an errno.
*
* Called by the ib_create_qp() core verbs function.
*/
struct ib_qp *hfi1_create_qp(struct ib_pd *ibpd,
struct ib_qp_init_attr *init_attr,
struct ib_udata *udata)
{
struct hfi1_qp *qp;
int err;
struct hfi1_swqe *swq = NULL;
struct hfi1_ibdev *dev;
struct hfi1_devdata *dd;
size_t sz;
size_t sg_list_sz;
struct ib_qp *ret;
if (init_attr->cap.max_send_sge > hfi1_max_sges ||
init_attr->cap.max_send_wr > hfi1_max_qp_wrs ||
init_attr->create_flags) {
ret = ERR_PTR(-EINVAL);
goto bail;
}
/* Check receive queue parameters if no SRQ is specified. */
if (!init_attr->srq) {
if (init_attr->cap.max_recv_sge > hfi1_max_sges ||
init_attr->cap.max_recv_wr > hfi1_max_qp_wrs) {
ret = ERR_PTR(-EINVAL);
goto bail;
}
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) {
ret = ERR_PTR(-EINVAL);
goto bail;
}
}
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) {
ret = ERR_PTR(-EINVAL);
goto bail;
}
case IB_QPT_UC:
case IB_QPT_RC:
case IB_QPT_UD:
sz = sizeof(struct hfi1_sge) *
init_attr->cap.max_send_sge +
sizeof(struct hfi1_swqe);
swq = vmalloc((init_attr->cap.max_send_wr + 1) * sz);
if (swq == NULL) {
ret = ERR_PTR(-ENOMEM);
goto bail;
}
sz = sizeof(*qp);
sg_list_sz = 0;
if (init_attr->srq) {
struct hfi1_srq *srq = to_isrq(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(sz + sg_list_sz, GFP_KERNEL);
if (!qp) {
ret = ERR_PTR(-ENOMEM);
goto bail_swq;
}
RCU_INIT_POINTER(qp->next, NULL);
qp->s_hdr = kzalloc(sizeof(*qp->s_hdr), GFP_KERNEL);
if (!qp->s_hdr) {
ret = ERR_PTR(-ENOMEM);
goto bail_qp;
}
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 hfi1_rwqe);
qp->r_rq.wq = vmalloc_user(sizeof(struct hfi1_rwq) +
qp->r_rq.size * sz);
if (!qp->r_rq.wq) {
ret = ERR_PTR(-ENOMEM);
goto bail_qp;
}
}
/*
* ib_create_qp() will initialize qp->ibqp
* except for qp->ibqp.qp_num.
*/
spin_lock_init(&qp->r_lock);
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_max_sge = init_attr->cap.max_send_sge;
if (init_attr->sq_sig_type == IB_SIGNAL_REQ_WR)
qp->s_flags = HFI1_S_SIGNAL_REQ_WR;
dev = to_idev(ibpd->device);
dd = dd_from_dev(dev);
err = alloc_qpn(dd, &dev->qp_dev->qpn_table, init_attr->qp_type,
init_attr->port_num);
if (err < 0) {
ret = ERR_PTR(err);
vfree(qp->r_rq.wq);
goto bail_qp;
}
qp->ibqp.qp_num = err;
qp->port_num = init_attr->port_num;
reset_qp(qp, init_attr->qp_type);
break;
default:
/* Don't support raw QPs */
ret = ERR_PTR(-ENOSYS);
goto bail;
}
init_attr->cap.max_inline_data = 0;
/*
* Return the address of the RWQ as the offset to mmap.
* See hfi1_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_ip;
}
} else {
u32 s = sizeof(struct hfi1_rwq) + qp->r_rq.size * sz;
qp->ip = hfi1_create_mmap_info(dev, s,
ibpd->uobject->context,
qp->r_rq.wq);
if (!qp->ip) {
ret = ERR_PTR(-ENOMEM);
goto bail_ip;
}
err = ib_copy_to_udata(udata, &(qp->ip->offset),
sizeof(qp->ip->offset));
if (err) {
ret = ERR_PTR(err);
goto bail_ip;
}
}
}
spin_lock(&dev->n_qps_lock);
if (dev->n_qps_allocated == hfi1_max_qps) {
spin_unlock(&dev->n_qps_lock);
ret = ERR_PTR(-ENOMEM);
goto bail_ip;
}
dev->n_qps_allocated++;
spin_unlock(&dev->n_qps_lock);
if (qp->ip) {
spin_lock_irq(&dev->pending_lock);
list_add(&qp->ip->pending_mmaps, &dev->pending_mmaps);
spin_unlock_irq(&dev->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 & OPCODE_QP_MASK;
break;
case IB_QPT_RC:
qp->allowed_ops = IB_OPCODE_RC_SEND_ONLY & OPCODE_QP_MASK;
break;
case IB_QPT_UC:
qp->allowed_ops = IB_OPCODE_UC_SEND_ONLY & OPCODE_QP_MASK;
break;
default:
ret = ERR_PTR(-EINVAL);
goto bail_ip;
}
goto bail;
bail_ip:
if (qp->ip)
kref_put(&qp->ip->ref, hfi1_release_mmap_info);
else
vfree(qp->r_rq.wq);
free_qpn(&dev->qp_dev->qpn_table, qp->ibqp.qp_num);
bail_qp:
kfree(qp->s_hdr);
kfree(qp);
bail_swq:
vfree(swq);
bail:
return ret;
}