/*
* Get one cq entry from c4iw and map it to openib.
*
* Returns:
* 0 cqe returned
* -ENODATA EMPTY;
* -EAGAIN caller must try again
* any other -errno fatal error
*/
static int c4iw_poll_cq_one(struct c4iw_cq *chp, struct ib_wc *wc)
{
struct c4iw_srq *srq = NULL;
struct c4iw_qp *qhp = NULL;
struct t4_cqe *rd_cqe;
int ret;
ret = t4_next_cqe(&chp->cq, &rd_cqe);
if (ret)
return ret;
qhp = get_qhp(chp->rhp, CQE_QPID(rd_cqe));
if (qhp) {
spin_lock(&qhp->lock);
srq = qhp->srq;
if (srq)
spin_lock(&srq->lock);
ret = __c4iw_poll_cq_one(chp, qhp, wc, srq);
spin_unlock(&qhp->lock);
if (srq)
spin_unlock(&srq->lock);
} else {
ret = __c4iw_poll_cq_one(chp, NULL, wc, NULL);
}
return ret;
}
/*
* Get one cq entry from c4iw and map it to openib.
*
* Returns:
* 0 cqe returned
* -ENODATA EMPTY;
* -EAGAIN caller must try again
* any other -errno fatal error
*/
static int c4iw_poll_cq_one(struct c4iw_cq *chp, struct ib_wc *wc)
{
struct c4iw_qp *qhp = NULL;
struct t4_cqe cqe = {0, 0}, *rd_cqe;
struct t4_wq *wq;
u32 credit = 0;
u8 cqe_flushed;
u64 cookie = 0;
int ret;
ret = t4_next_cqe(&chp->cq, &rd_cqe);
if (ret)
return ret;
qhp = get_qhp(chp->rhp, CQE_QPID(rd_cqe));
if (!qhp)
wq = NULL;
else {
spin_lock(&qhp->lock);
wq = &(qhp->wq);
}
ret = poll_cq(wq, &(chp->cq), &cqe, &cqe_flushed, &cookie, &credit);
if (ret)
goto out;
wc->wr_id = cookie;
wc->qp = &qhp->ibqp;
wc->vendor_err = CQE_STATUS(&cqe);
wc->wc_flags = 0;
PDBG("%s qpid 0x%x type %d opcode %d status 0x%x len %u wrid hi 0x%x "
"lo 0x%x cookie 0x%llx\n", __func__, CQE_QPID(&cqe),
CQE_TYPE(&cqe), CQE_OPCODE(&cqe), CQE_STATUS(&cqe), CQE_LEN(&cqe),
CQE_WRID_HI(&cqe), CQE_WRID_LOW(&cqe), (unsigned long long)cookie);
if (CQE_TYPE(&cqe) == 0) {
if (!CQE_STATUS(&cqe))
wc->byte_len = CQE_LEN(&cqe);
else
wc->byte_len = 0;
wc->opcode = IB_WC_RECV;
if (CQE_OPCODE(&cqe) == FW_RI_SEND_WITH_INV ||
CQE_OPCODE(&cqe) == FW_RI_SEND_WITH_SE_INV) {
wc->ex.invalidate_rkey = CQE_WRID_STAG(&cqe);
wc->wc_flags |= IB_WC_WITH_INVALIDATE;
}
} else {
switch (CQE_OPCODE(&cqe)) {
case FW_RI_RDMA_WRITE:
wc->opcode = IB_WC_RDMA_WRITE;
break;
case FW_RI_READ_REQ:
wc->opcode = IB_WC_RDMA_READ;
wc->byte_len = CQE_LEN(&cqe);
break;
case FW_RI_SEND_WITH_INV:
case FW_RI_SEND_WITH_SE_INV:
wc->opcode = IB_WC_SEND;
wc->wc_flags |= IB_WC_WITH_INVALIDATE;
break;
case FW_RI_SEND:
case FW_RI_SEND_WITH_SE:
wc->opcode = IB_WC_SEND;
break;
case FW_RI_BIND_MW:
wc->opcode = IB_WC_BIND_MW;
break;
case FW_RI_LOCAL_INV:
wc->opcode = IB_WC_LOCAL_INV;
break;
case FW_RI_FAST_REGISTER:
wc->opcode = IB_WC_FAST_REG_MR;
break;
default:
printk(KERN_ERR MOD "Unexpected opcode %d "
"in the CQE received for QPID=0x%0x\n",
CQE_OPCODE(&cqe), CQE_QPID(&cqe));
ret = -EINVAL;
goto out;
}
}
if (cqe_flushed)
wc->status = IB_WC_WR_FLUSH_ERR;
else {
switch (CQE_STATUS(&cqe)) {
case T4_ERR_SUCCESS:
wc->status = IB_WC_SUCCESS;
break;
case T4_ERR_STAG:
wc->status = IB_WC_LOC_ACCESS_ERR;
break;
case T4_ERR_PDID:
wc->status = IB_WC_LOC_PROT_ERR;
break;
case T4_ERR_QPID:
case T4_ERR_ACCESS:
wc->status = IB_WC_LOC_ACCESS_ERR;
break;
case T4_ERR_WRAP:
wc->status = IB_WC_GENERAL_ERR;
break;
case T4_ERR_BOUND:
wc->status = IB_WC_LOC_LEN_ERR;
break;
case T4_ERR_INVALIDATE_SHARED_MR:
case T4_ERR_INVALIDATE_MR_WITH_MW_BOUND:
wc->status = IB_WC_MW_BIND_ERR;
break;
case T4_ERR_CRC:
case T4_ERR_MARKER:
case T4_ERR_PDU_LEN_ERR:
case T4_ERR_OUT_OF_RQE:
case T4_ERR_DDP_VERSION:
case T4_ERR_RDMA_VERSION:
case T4_ERR_DDP_QUEUE_NUM:
case T4_ERR_MSN:
case T4_ERR_TBIT:
case T4_ERR_MO:
case T4_ERR_MSN_RANGE:
case T4_ERR_IRD_OVERFLOW:
case T4_ERR_OPCODE:
case T4_ERR_INTERNAL_ERR:
wc->status = IB_WC_FATAL_ERR;
break;
case T4_ERR_SWFLUSH:
wc->status = IB_WC_WR_FLUSH_ERR;
break;
default:
printk(KERN_ERR MOD
"Unexpected cqe_status 0x%x for QPID=0x%0x\n",
CQE_STATUS(&cqe), CQE_QPID(&cqe));
ret = -EINVAL;
}
}
out:
if (wq)
spin_unlock(&qhp->lock);
return ret;
}
/*
* 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;
}
/*
* 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;
}
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;
}