static int cqe_completes_wr(struct t4_cqe *cqe, struct t4_wq *wq)
{
if (CQE_OPCODE(cqe) == FW_RI_TERMINATE)
return 0;
if ((CQE_OPCODE(cqe) == FW_RI_RDMA_WRITE) && RQ_TYPE(cqe))
return 0;
if ((CQE_OPCODE(cqe) == FW_RI_READ_RESP) && SQ_TYPE(cqe))
return 0;
if (CQE_SEND_OPCODE(cqe) && RQ_TYPE(cqe) && t4_rq_empty(wq))
return 0;
return 1;
}
static int cqe_completes_wr(struct t4_cqe *cqe, struct t4_wq *wq)
{
if (DRAIN_CQE(cqe)) {
WARN_ONCE(1, "Unexpected DRAIN CQE qp id %u!\n", wq->sq.qid);
return 0;
}
if (CQE_OPCODE(cqe) == FW_RI_TERMINATE)
return 0;
if ((CQE_OPCODE(cqe) == FW_RI_RDMA_WRITE) && RQ_TYPE(cqe))
return 0;
if ((CQE_OPCODE(cqe) == FW_RI_READ_RESP) && SQ_TYPE(cqe))
return 0;
if (CQE_SEND_OPCODE(cqe) && RQ_TYPE(cqe) && t4_rq_empty(wq))
return 0;
return 1;
}
void c4iw_count_rcqes(struct t4_cq *cq, struct t4_wq *wq, int *count)
{
struct t4_cqe *cqe;
u32 ptr;
*count = 0;
PDBG("%s count zero %d\n", __func__, *count);
ptr = cq->sw_cidx;
while (ptr != cq->sw_pidx) {
cqe = &cq->sw_queue[ptr];
if (RQ_TYPE(cqe) && (CQE_OPCODE(cqe) != FW_RI_READ_RESP) &&
(CQE_QPID(cqe) == wq->rq.qid) && cqe_completes_wr(cqe, wq))
(*count)++;
if (++ptr == cq->size)
ptr = 0;
}
PDBG("%s cq %p count %d\n", __func__, cq, *count);
}
/*
* poll_cq
*
* Caller must:
* check the validity of the first CQE,
* supply the wq assicated with the qpid.
*
* credit: cq credit to return to sge.
* cqe_flushed: 1 iff the CQE is flushed.
* cqe: copy of the polled CQE.
*
* return value:
* 0 CQE returned ok.
* -EAGAIN CQE skipped, try again.
* -EOVERFLOW CQ overflow detected.
*/
static int poll_cq(struct t4_wq *wq, struct t4_cq *cq, struct t4_cqe *cqe,
u8 *cqe_flushed, u64 *cookie, u32 *credit)
{
int ret = 0;
struct t4_cqe *hw_cqe, read_cqe;
*cqe_flushed = 0;
*credit = 0;
ret = t4_next_cqe(cq, &hw_cqe);
if (ret)
return ret;
PDBG("%s CQE OVF %u qpid 0x%0x genbit %u type %u status 0x%0x"
" opcode 0x%0x len 0x%0x wrid_hi_stag 0x%x wrid_low_msn 0x%x\n",
__func__, CQE_OVFBIT(hw_cqe), CQE_QPID(hw_cqe),
CQE_GENBIT(hw_cqe), CQE_TYPE(hw_cqe), CQE_STATUS(hw_cqe),
CQE_OPCODE(hw_cqe), CQE_LEN(hw_cqe), CQE_WRID_HI(hw_cqe),
CQE_WRID_LOW(hw_cqe));
/*
* skip cqe's not affiliated with a QP.
*/
if (wq == NULL) {
ret = -EAGAIN;
goto skip_cqe;
}
/*
* Gotta tweak READ completions:
* 1) the cqe doesn't contain the sq_wptr from the wr.
* 2) opcode not reflected from the wr.
* 3) read_len not reflected from the wr.
* 4) cq_type is RQ_TYPE not SQ_TYPE.
*/
if (RQ_TYPE(hw_cqe) && (CQE_OPCODE(hw_cqe) == FW_RI_READ_RESP)) {
/*
* If this is an unsolicited read response, then the read
* was generated by the kernel driver as part of peer-2-peer
* connection setup. So ignore the completion.
*/
if (!wq->sq.oldest_read) {
if (CQE_STATUS(hw_cqe))
t4_set_wq_in_error(wq);
ret = -EAGAIN;
goto skip_cqe;
}
/*
* Don't write to the HWCQ, so create a new read req CQE
* in local memory.
*/
create_read_req_cqe(wq, hw_cqe, &read_cqe);
hw_cqe = &read_cqe;
advance_oldest_read(wq);
}
if (CQE_STATUS(hw_cqe) || t4_wq_in_error(wq)) {
*cqe_flushed = t4_wq_in_error(wq);
t4_set_wq_in_error(wq);
goto proc_cqe;
}
if (CQE_OPCODE(hw_cqe) == FW_RI_TERMINATE) {
ret = -EAGAIN;
goto skip_cqe;
}
/*
* RECV completion.
*/
if (RQ_TYPE(hw_cqe)) {
/*
* HW only validates 4 bits of MSN. So we must validate that
* the MSN in the SEND is the next expected MSN. If its not,
* then we complete this with T4_ERR_MSN and mark the wq in
* error.
*/
if (t4_rq_empty(wq)) {
t4_set_wq_in_error(wq);
ret = -EAGAIN;
goto skip_cqe;
}
if (unlikely((CQE_WRID_MSN(hw_cqe) != (wq->rq.msn)))) {
t4_set_wq_in_error(wq);
hw_cqe->header |= htonl(V_CQE_STATUS(T4_ERR_MSN));
goto proc_cqe;
}
goto proc_cqe;
}
/*
* If we get here its a send completion.
*
* Handle out of order completion. These get stuffed
* in the SW SQ. Then the SW SQ is walked to move any
* now in-order completions into the SW CQ. This handles
* 2 cases:
* 1) reaping unsignaled WRs when the first subsequent
* signaled WR is completed.
* 2) out of order read completions.
*/
if (!SW_CQE(hw_cqe) && (CQE_WRID_SQ_IDX(hw_cqe) != wq->sq.cidx)) {
struct t4_swsqe *swsqe;
PDBG("%s out of order completion going in sw_sq at idx %u\n",
__func__, CQE_WRID_SQ_IDX(hw_cqe));
swsqe = &wq->sq.sw_sq[CQE_WRID_SQ_IDX(hw_cqe)];
swsqe->cqe = *hw_cqe;
swsqe->complete = 1;
ret = -EAGAIN;
goto flush_wq;
}
proc_cqe:
*cqe = *hw_cqe;
/*
* Reap the associated WR(s) that are freed up with this
* completion.
*/
if (SQ_TYPE(hw_cqe)) {
wq->sq.cidx = CQE_WRID_SQ_IDX(hw_cqe);
PDBG("%s completing sq idx %u\n", __func__, wq->sq.cidx);
*cookie = wq->sq.sw_sq[wq->sq.cidx].wr_id;
t4_sq_consume(wq);
} else {
PDBG("%s completing rq idx %u\n", __func__, wq->rq.cidx);
*cookie = wq->rq.sw_rq[wq->rq.cidx].wr_id;
BUG_ON(t4_rq_empty(wq));
t4_rq_consume(wq);
}
flush_wq:
/*
* Flush any completed cqes that are now in-order.
*/
flush_completed_wrs(wq, cq);
skip_cqe:
if (SW_CQE(hw_cqe)) {
PDBG("%s cq %p cqid 0x%x skip sw cqe cidx %u\n",
__func__, cq, cq->cqid, cq->sw_cidx);
t4_swcq_consume(cq);
} else {
PDBG("%s cq %p cqid 0x%x skip hw cqe cidx %u\n",
__func__, cq, cq->cqid, cq->cidx);
t4_hwcq_consume(cq);
}
return ret;
}
void c4iw_ev_dispatch(struct c4iw_dev *dev, struct t4_cqe *err_cqe)
{
struct c4iw_cq *chp;
struct c4iw_qp *qhp;
u32 cqid;
spin_lock(&dev->lock);
qhp = get_qhp(dev, CQE_QPID(err_cqe));
if (!qhp) {
printk(KERN_ERR MOD "BAD AE qpid 0x%x opcode %d "
"status 0x%x type %d wrid.hi 0x%x wrid.lo 0x%x\n",
CQE_QPID(err_cqe),
CQE_OPCODE(err_cqe), CQE_STATUS(err_cqe),
CQE_TYPE(err_cqe), CQE_WRID_HI(err_cqe),
CQE_WRID_LOW(err_cqe));
spin_unlock(&dev->lock);
goto out;
}
if (SQ_TYPE(err_cqe))
cqid = qhp->attr.scq;
else
cqid = qhp->attr.rcq;
chp = get_chp(dev, cqid);
if (!chp) {
printk(KERN_ERR MOD "BAD AE cqid 0x%x qpid 0x%x opcode %d "
"status 0x%x type %d wrid.hi 0x%x wrid.lo 0x%x\n",
cqid, CQE_QPID(err_cqe),
CQE_OPCODE(err_cqe), CQE_STATUS(err_cqe),
CQE_TYPE(err_cqe), CQE_WRID_HI(err_cqe),
CQE_WRID_LOW(err_cqe));
spin_unlock(&dev->lock);
goto out;
}
c4iw_qp_add_ref(&qhp->ibqp);
atomic_inc(&chp->refcnt);
spin_unlock(&dev->lock);
/* Bad incoming write */
if (RQ_TYPE(err_cqe) &&
(CQE_OPCODE(err_cqe) == FW_RI_RDMA_WRITE)) {
post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_QP_REQ_ERR);
goto done;
}
switch (CQE_STATUS(err_cqe)) {
/* Completion Events */
case T4_ERR_SUCCESS:
printk(KERN_ERR MOD "AE with status 0!\n");
break;
case T4_ERR_STAG:
case T4_ERR_PDID:
case T4_ERR_QPID:
case T4_ERR_ACCESS:
case T4_ERR_WRAP:
case T4_ERR_BOUND:
case T4_ERR_INVALIDATE_SHARED_MR:
case T4_ERR_INVALIDATE_MR_WITH_MW_BOUND:
post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_QP_ACCESS_ERR);
break;
/* Device Fatal Errors */
case T4_ERR_ECC:
case T4_ERR_ECC_PSTAG:
case T4_ERR_INTERNAL_ERR:
post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_DEVICE_FATAL);
break;
/* QP Fatal Errors */
case T4_ERR_OUT_OF_RQE:
case T4_ERR_PBL_ADDR_BOUND:
case T4_ERR_CRC:
case T4_ERR_MARKER:
case T4_ERR_PDU_LEN_ERR:
case T4_ERR_DDP_VERSION:
case T4_ERR_RDMA_VERSION:
case T4_ERR_OPCODE:
case T4_ERR_DDP_QUEUE_NUM:
case T4_ERR_MSN:
case T4_ERR_TBIT:
case T4_ERR_MO:
case T4_ERR_MSN_GAP:
case T4_ERR_MSN_RANGE:
case T4_ERR_RQE_ADDR_BOUND:
case T4_ERR_IRD_OVERFLOW:
post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_QP_FATAL);
break;
default:
printk(KERN_ERR MOD "Unknown T4 status 0x%x QPID 0x%x\n",
CQE_STATUS(err_cqe), qhp->wq.sq.qid);
post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_QP_FATAL);
break;
}
done:
if (atomic_dec_and_test(&chp->refcnt))
wake_up(&chp->wait);
c4iw_qp_rem_ref(&qhp->ibqp);
out:
return;
}
/*
* poll_cq
*
* Caller must:
* check the validity of the first CQE,
* supply the wq assicated with the qpid.
*
* credit: cq credit to return to sge.
* cqe_flushed: 1 iff the CQE is flushed.
* cqe: copy of the polled CQE.
*
* return value:
* 0 CQE returned ok.
* -EAGAIN CQE skipped, try again.
* -EOVERFLOW CQ overflow detected.
*/
static int poll_cq(struct t4_wq *wq, struct t4_cq *cq, struct t4_cqe *cqe,
u8 *cqe_flushed, u64 *cookie, u32 *credit,
struct t4_srq *srq)
{
int ret = 0;
struct t4_cqe *hw_cqe, read_cqe;
*cqe_flushed = 0;
*credit = 0;
ret = t4_next_cqe(cq, &hw_cqe);
if (ret)
return ret;
pr_debug("CQE OVF %u qpid 0x%0x genbit %u type %u status 0x%0x opcode 0x%0x len 0x%0x wrid_hi_stag 0x%x wrid_low_msn 0x%x\n",
CQE_OVFBIT(hw_cqe), CQE_QPID(hw_cqe),
CQE_GENBIT(hw_cqe), CQE_TYPE(hw_cqe), CQE_STATUS(hw_cqe),
CQE_OPCODE(hw_cqe), CQE_LEN(hw_cqe), CQE_WRID_HI(hw_cqe),
CQE_WRID_LOW(hw_cqe));
/*
* skip cqe's not affiliated with a QP.
*/
if (wq == NULL) {
ret = -EAGAIN;
goto skip_cqe;
}
/*
* skip hw cqe's if the wq is flushed.
*/
if (wq->flushed && !SW_CQE(hw_cqe)) {
ret = -EAGAIN;
goto skip_cqe;
}
/*
* skip TERMINATE cqes...
*/
if (CQE_OPCODE(hw_cqe) == FW_RI_TERMINATE) {
ret = -EAGAIN;
goto skip_cqe;
}
/*
* Special cqe for drain WR completions...
*/
if (DRAIN_CQE(hw_cqe)) {
*cookie = CQE_DRAIN_COOKIE(hw_cqe);
*cqe = *hw_cqe;
goto skip_cqe;
}
/*
* Gotta tweak READ completions:
* 1) the cqe doesn't contain the sq_wptr from the wr.
* 2) opcode not reflected from the wr.
* 3) read_len not reflected from the wr.
* 4) cq_type is RQ_TYPE not SQ_TYPE.
*/
if (RQ_TYPE(hw_cqe) && (CQE_OPCODE(hw_cqe) == FW_RI_READ_RESP)) {
/* If we have reached here because of async
* event or other error, and have egress error
* then drop
*/
if (CQE_TYPE(hw_cqe) == 1) {
if (CQE_STATUS(hw_cqe))
t4_set_wq_in_error(wq, 0);
ret = -EAGAIN;
goto skip_cqe;
}
/* If this is an unsolicited read response, then the read
* was generated by the kernel driver as part of peer-2-peer
* connection setup. So ignore the completion.
*/
if (CQE_WRID_STAG(hw_cqe) == 1) {
if (CQE_STATUS(hw_cqe))
t4_set_wq_in_error(wq, 0);
ret = -EAGAIN;
goto skip_cqe;
}
/*
* Eat completions for unsignaled read WRs.
*/
if (!wq->sq.oldest_read->signaled) {
advance_oldest_read(wq);
ret = -EAGAIN;
goto skip_cqe;
}
/*
* Don't write to the HWCQ, so create a new read req CQE
* in local memory.
*/
create_read_req_cqe(wq, hw_cqe, &read_cqe);
hw_cqe = &read_cqe;
advance_oldest_read(wq);
}
if (CQE_STATUS(hw_cqe) || t4_wq_in_error(wq)) {
*cqe_flushed = (CQE_STATUS(hw_cqe) == T4_ERR_SWFLUSH);
t4_set_wq_in_error(wq, 0);
}
/*
* RECV completion.
*/
if (RQ_TYPE(hw_cqe)) {
/*
* HW only validates 4 bits of MSN. So we must validate that
* the MSN in the SEND is the next expected MSN. If its not,
* then we complete this with T4_ERR_MSN and mark the wq in
* error.
*/
if (unlikely(!CQE_STATUS(hw_cqe) &&
CQE_WRID_MSN(hw_cqe) != wq->rq.msn)) {
t4_set_wq_in_error(wq, 0);
hw_cqe->header |= cpu_to_be32(CQE_STATUS_V(T4_ERR_MSN));
}
goto proc_cqe;
}
/*
* If we get here its a send completion.
*
* Handle out of order completion. These get stuffed
* in the SW SQ. Then the SW SQ is walked to move any
* now in-order completions into the SW CQ. This handles
* 2 cases:
* 1) reaping unsignaled WRs when the first subsequent
* signaled WR is completed.
* 2) out of order read completions.
*/
if (!SW_CQE(hw_cqe) && (CQE_WRID_SQ_IDX(hw_cqe) != wq->sq.cidx)) {
struct t4_swsqe *swsqe;
pr_debug("out of order completion going in sw_sq at idx %u\n",
CQE_WRID_SQ_IDX(hw_cqe));
swsqe = &wq->sq.sw_sq[CQE_WRID_SQ_IDX(hw_cqe)];
swsqe->cqe = *hw_cqe;
swsqe->complete = 1;
ret = -EAGAIN;
goto flush_wq;
}
proc_cqe:
*cqe = *hw_cqe;
/*
* Reap the associated WR(s) that are freed up with this
* completion.
*/
if (SQ_TYPE(hw_cqe)) {
int idx = CQE_WRID_SQ_IDX(hw_cqe);
/*
* Account for any unsignaled completions completed by
* this signaled completion. In this case, cidx points
* to the first unsignaled one, and idx points to the
* signaled one. So adjust in_use based on this delta.
* if this is not completing any unsigned wrs, then the
* delta will be 0. Handle wrapping also!
*/
if (idx < wq->sq.cidx)
wq->sq.in_use -= wq->sq.size + idx - wq->sq.cidx;
else
wq->sq.in_use -= idx - wq->sq.cidx;
wq->sq.cidx = (uint16_t)idx;
pr_debug("completing sq idx %u\n", wq->sq.cidx);
*cookie = wq->sq.sw_sq[wq->sq.cidx].wr_id;
if (c4iw_wr_log)
c4iw_log_wr_stats(wq, hw_cqe);
t4_sq_consume(wq);
} else {
if (!srq) {
pr_debug("completing rq idx %u\n", wq->rq.cidx);
*cookie = wq->rq.sw_rq[wq->rq.cidx].wr_id;
if (c4iw_wr_log)
c4iw_log_wr_stats(wq, hw_cqe);
t4_rq_consume(wq);
} else {
*cookie = reap_srq_cqe(hw_cqe, srq);
}
wq->rq.msn++;
goto skip_cqe;
}
flush_wq:
/*
* Flush any completed cqes that are now in-order.
*/
flush_completed_wrs(wq, cq);
skip_cqe:
if (SW_CQE(hw_cqe)) {
pr_debug("cq %p cqid 0x%x skip sw cqe cidx %u\n",
cq, cq->cqid, cq->sw_cidx);
t4_swcq_consume(cq);
} else {
pr_debug("cq %p cqid 0x%x skip hw cqe cidx %u\n",
cq, cq->cqid, cq->cidx);
t4_hwcq_consume(cq);
}
return ret;
}
void iwch_ev_dispatch(struct cxio_rdev *rdev_p, struct sk_buff *skb)
{
struct iwch_dev *rnicp;
struct respQ_msg_t *rsp_msg = (struct respQ_msg_t *) skb->data;
struct iwch_cq *chp;
struct iwch_qp *qhp;
u32 cqid = RSPQ_CQID(rsp_msg);
rnicp = (struct iwch_dev *) rdev_p->ulp;
spin_lock(&rnicp->lock);
chp = get_chp(rnicp, cqid);
qhp = get_qhp(rnicp, CQE_QPID(rsp_msg->cqe));
if (!chp || !qhp) {
printk(KERN_ERR MOD "BAD AE cqid 0x%x qpid 0x%x opcode %d "
"status 0x%x type %d wrid.hi 0x%x wrid.lo 0x%x \n",
cqid, CQE_QPID(rsp_msg->cqe),
CQE_OPCODE(rsp_msg->cqe), CQE_STATUS(rsp_msg->cqe),
CQE_TYPE(rsp_msg->cqe), CQE_WRID_HI(rsp_msg->cqe),
CQE_WRID_LOW(rsp_msg->cqe));
spin_unlock(&rnicp->lock);
goto out;
}
iwch_qp_add_ref(&qhp->ibqp);
atomic_inc(&chp->refcnt);
spin_unlock(&rnicp->lock);
/*
* 1) completion of our sending a TERMINATE.
* 2) incoming TERMINATE message.
*/
if ((CQE_OPCODE(rsp_msg->cqe) == T3_TERMINATE) &&
(CQE_STATUS(rsp_msg->cqe) == 0)) {
if (SQ_TYPE(rsp_msg->cqe)) {
PDBG("%s QPID 0x%x ep %p disconnecting\n",
__func__, qhp->wq.qpid, qhp->ep);
iwch_ep_disconnect(qhp->ep, 0, GFP_ATOMIC);
} else {
PDBG("%s post REQ_ERR AE QPID 0x%x\n", __func__,
qhp->wq.qpid);
post_qp_event(rnicp, chp, rsp_msg,
IB_EVENT_QP_REQ_ERR, 0);
iwch_ep_disconnect(qhp->ep, 0, GFP_ATOMIC);
}
goto done;
}
/* Bad incoming Read request */
if (SQ_TYPE(rsp_msg->cqe) &&
(CQE_OPCODE(rsp_msg->cqe) == T3_READ_RESP)) {
post_qp_event(rnicp, chp, rsp_msg, IB_EVENT_QP_REQ_ERR, 1);
goto done;
}
/* Bad incoming write */
if (RQ_TYPE(rsp_msg->cqe) &&
(CQE_OPCODE(rsp_msg->cqe) == T3_RDMA_WRITE)) {
post_qp_event(rnicp, chp, rsp_msg, IB_EVENT_QP_REQ_ERR, 1);
goto done;
}
switch (CQE_STATUS(rsp_msg->cqe)) {
/* Completion Events */
case TPT_ERR_SUCCESS:
/*
* Confirm the destination entry if this is a RECV completion.
*/
if (qhp->ep && SQ_TYPE(rsp_msg->cqe))
dst_confirm(qhp->ep->dst);
(*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
break;
case TPT_ERR_STAG:
case TPT_ERR_PDID:
case TPT_ERR_QPID:
case TPT_ERR_ACCESS:
case TPT_ERR_WRAP:
case TPT_ERR_BOUND:
case TPT_ERR_INVALIDATE_SHARED_MR:
case TPT_ERR_INVALIDATE_MR_WITH_MW_BOUND:
post_qp_event(rnicp, chp, rsp_msg, IB_EVENT_QP_ACCESS_ERR, 1);
break;
/* Device Fatal Errors */
case TPT_ERR_ECC:
case TPT_ERR_ECC_PSTAG:
case TPT_ERR_INTERNAL_ERR:
post_qp_event(rnicp, chp, rsp_msg, IB_EVENT_DEVICE_FATAL, 1);
break;
/* QP Fatal Errors */
case TPT_ERR_OUT_OF_RQE:
case TPT_ERR_PBL_ADDR_BOUND:
case TPT_ERR_CRC:
case TPT_ERR_MARKER:
case TPT_ERR_PDU_LEN_ERR:
case TPT_ERR_DDP_VERSION:
case TPT_ERR_RDMA_VERSION:
case TPT_ERR_OPCODE:
case TPT_ERR_DDP_QUEUE_NUM:
case TPT_ERR_MSN:
case TPT_ERR_TBIT:
case TPT_ERR_MO:
case TPT_ERR_MSN_GAP:
case TPT_ERR_MSN_RANGE:
case TPT_ERR_RQE_ADDR_BOUND:
case TPT_ERR_IRD_OVERFLOW:
post_qp_event(rnicp, chp, rsp_msg, IB_EVENT_QP_FATAL, 1);
break;
default:
printk(KERN_ERR MOD "Unknown T3 status 0x%x QPID 0x%x\n",
CQE_STATUS(rsp_msg->cqe), qhp->wq.qpid);
post_qp_event(rnicp, chp, rsp_msg, IB_EVENT_QP_FATAL, 1);
break;
}
done:
if (atomic_dec_and_test(&chp->refcnt))
wake_up(&chp->wait);
iwch_qp_rem_ref(&qhp->ibqp);
out:
dev_kfree_skb_irq(skb);
}
/*
* do_local_work
*
* Any processing errors are placed in the 'rq'
* structure to be reported back to the kernel.
* It may be pointless for this function to
* return an int.
*
* Returns: 0 on success, -EXXX on failure
*/
static int do_local_work(void *data)
{
int r;
struct clog_request *rq;
struct dm_ulog_request *u_rq = NULL;
r = kernel_recv(&rq);
if (r)
return r;
if (!rq)
return 0;
u_rq = &rq->u_rq;
LOG_DBG("[%s] Request from kernel received: [%s/%u]",
SHORT_UUID(u_rq->uuid), RQ_TYPE(u_rq->request_type),
u_rq->seq);
switch (u_rq->request_type) {
case DM_ULOG_CTR:
case DM_ULOG_DTR:
case DM_ULOG_GET_REGION_SIZE:
case DM_ULOG_IN_SYNC:
case DM_ULOG_GET_SYNC_COUNT:
case DM_ULOG_STATUS_INFO:
case DM_ULOG_STATUS_TABLE:
case DM_ULOG_PRESUSPEND:
/* We do not specify ourselves as server here */
r = do_request(rq, 0);
if (r)
LOG_DBG("Returning failed request to kernel [%s]",
RQ_TYPE(u_rq->request_type));
r = kernel_send(u_rq);
if (r)
LOG_ERROR("Failed to respond to kernel [%s]",
RQ_TYPE(u_rq->request_type));
break;
case DM_ULOG_RESUME:
/*
* Resume is a special case that requires a local
* component to join the CPG, and a cluster component
* to handle the request.
*/
r = local_resume(u_rq);
if (r) {
LOG_DBG("Returning failed request to kernel [%s]",
RQ_TYPE(u_rq->request_type));
r = kernel_send(u_rq);
if (r)
LOG_ERROR("Failed to respond to kernel [%s]",
RQ_TYPE(u_rq->request_type));
break;
}
/* ELSE, fall through */
case DM_ULOG_IS_CLEAN:
case DM_ULOG_FLUSH:
case DM_ULOG_MARK_REGION:
case DM_ULOG_GET_RESYNC_WORK:
case DM_ULOG_SET_REGION_SYNC:
case DM_ULOG_IS_REMOTE_RECOVERING:
case DM_ULOG_POSTSUSPEND:
r = cluster_send(rq);
if (r) {
u_rq->data_size = 0;
u_rq->error = r;
kernel_send(u_rq);
}
break;
case DM_ULOG_CLEAR_REGION:
r = kernel_ack(u_rq->seq, 0);
r = cluster_send(rq);
if (r) {
/*
* FIXME: store error for delivery on flush
* This would allow us to optimize MARK_REGION
* too.
*/
}
break;
default:
LOG_ERROR("Invalid log request received (%u), ignoring.",
u_rq->request_type);
return 0;
}
if (r && !u_rq->error)
u_rq->error = r;
return r;
}
static int poll_cq(struct t4_wq *wq, struct t4_cq *cq, struct t4_cqe *cqe,
u8 *cqe_flushed, u64 *cookie, u32 *credit)
{
int ret = 0;
struct t4_cqe *hw_cqe, read_cqe;
*cqe_flushed = 0;
*credit = 0;
ret = t4_next_cqe(cq, &hw_cqe);
if (ret)
return ret;
PDBG("%s CQE OVF %u qpid 0x%0x genbit %u type %u status 0x%0x"
" opcode 0x%0x len 0x%0x wrid_hi_stag 0x%x wrid_low_msn 0x%x\n",
__func__, CQE_OVFBIT(hw_cqe), CQE_QPID(hw_cqe),
CQE_GENBIT(hw_cqe), CQE_TYPE(hw_cqe), CQE_STATUS(hw_cqe),
CQE_OPCODE(hw_cqe), CQE_LEN(hw_cqe), CQE_WRID_HI(hw_cqe),
CQE_WRID_LOW(hw_cqe));
/*
*/
if (wq == NULL) {
ret = -EAGAIN;
goto skip_cqe;
}
/*
*/
if (RQ_TYPE(hw_cqe) && (CQE_OPCODE(hw_cqe) == FW_RI_READ_RESP)) {
/*
*/
if (!wq->sq.oldest_read) {
if (CQE_STATUS(hw_cqe))
t4_set_wq_in_error(wq);
ret = -EAGAIN;
goto skip_cqe;
}
/*
*/
create_read_req_cqe(wq, hw_cqe, &read_cqe);
hw_cqe = &read_cqe;
advance_oldest_read(wq);
}
if (CQE_STATUS(hw_cqe) || t4_wq_in_error(wq)) {
*cqe_flushed = t4_wq_in_error(wq);
t4_set_wq_in_error(wq);
goto proc_cqe;
}
if (CQE_OPCODE(hw_cqe) == FW_RI_TERMINATE) {
ret = -EAGAIN;
goto skip_cqe;
}
/*
*/
if (RQ_TYPE(hw_cqe)) {
/*
*/
if (t4_rq_empty(wq)) {
t4_set_wq_in_error(wq);
ret = -EAGAIN;
goto skip_cqe;
}
if (unlikely((CQE_WRID_MSN(hw_cqe) != (wq->rq.msn)))) {
t4_set_wq_in_error(wq);
hw_cqe->header |= htonl(V_CQE_STATUS(T4_ERR_MSN));
goto proc_cqe;
}
goto proc_cqe;
}
/*
*/
if (!SW_CQE(hw_cqe) && (CQE_WRID_SQ_IDX(hw_cqe) != wq->sq.cidx)) {
struct t4_swsqe *swsqe;
PDBG("%s out of order completion going in sw_sq at idx %u\n",
__func__, CQE_WRID_SQ_IDX(hw_cqe));
swsqe = &wq->sq.sw_sq[CQE_WRID_SQ_IDX(hw_cqe)];
swsqe->cqe = *hw_cqe;
swsqe->complete = 1;
ret = -EAGAIN;
goto flush_wq;
}
proc_cqe:
*cqe = *hw_cqe;
/*
*/
if (SQ_TYPE(hw_cqe)) {
wq->sq.cidx = CQE_WRID_SQ_IDX(hw_cqe);
PDBG("%s completing sq idx %u\n", __func__, wq->sq.cidx);
*cookie = wq->sq.sw_sq[wq->sq.cidx].wr_id;
t4_sq_consume(wq);
} else {
PDBG("%s completing rq idx %u\n", __func__, wq->rq.cidx);
*cookie = wq->rq.sw_rq[wq->rq.cidx].wr_id;
BUG_ON(t4_rq_empty(wq));
t4_rq_consume(wq);
}
flush_wq:
/*
*/
flush_completed_wrs(wq, cq);
skip_cqe:
if (SW_CQE(hw_cqe)) {
PDBG("%s cq %p cqid 0x%x skip sw cqe cidx %u\n",
__func__, cq, cq->cqid, cq->sw_cidx);
t4_swcq_consume(cq);
} else {
PDBG("%s cq %p cqid 0x%x skip hw cqe cidx %u\n",
__func__, cq, cq->cqid, cq->cidx);
t4_hwcq_consume(cq);
}
return ret;
}