[PATCH for-rc 1/2] RDMA/irdma: Report correct WC errors

[PATCH for-rc 1/2] RDMA/irdma: Report correct WC errors

[Shiraz Saleem]

Return IBV_WC_REM_OP_ERR for responder QP errors instead of
IBV_WC_REM_ACCESS_ERR.

Return IBV_WC_LOC_QP_OP_ERR for errors detected on the SQ with bad opcodes

Fixes: 44d9e52977a1 ("RDMA/irdma: Implement device initialization definitions")
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
---
 drivers/infiniband/hw/irdma/hw.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/irdma/hw.c b/drivers/infiniband/hw/irdma/hw.c
index 4108dca..1bae1dc 100644
--- a/drivers/infiniband/hw/irdma/hw.c
+++ b/drivers/infiniband/hw/irdma/hw.c
@@ -146,6 +146,7 @@ static void irdma_set_flush_fields(struct irdma_sc_qp *qp,
 		qp->flush_code = FLUSH_PROT_ERR;
 		break;
 	case IRDMA_AE_AMP_BAD_QP:
+	case IRDMA_AE_WQE_UNEXPECTED_OPCODE:
 		qp->flush_code = FLUSH_LOC_QP_OP_ERR;
 		break;
 	case IRDMA_AE_AMP_BAD_STAG_KEY:
@@ -156,7 +157,6 @@ static void irdma_set_flush_fields(struct irdma_sc_qp *qp,
 	case IRDMA_AE_PRIV_OPERATION_DENIED:
 	case IRDMA_AE_IB_INVALID_REQUEST:
 	case IRDMA_AE_IB_REMOTE_ACCESS_ERROR:
-	case IRDMA_AE_IB_REMOTE_OP_ERROR:
 		qp->flush_code = FLUSH_REM_ACCESS_ERR;
 		qp->event_type = IRDMA_QP_EVENT_ACCESS_ERR;
 		break;
@@ -184,6 +184,9 @@ static void irdma_set_flush_fields(struct irdma_sc_qp *qp,
 	case IRDMA_AE_AMP_MWBIND_INVALID_BOUNDS:
 		qp->flush_code = FLUSH_MW_BIND_ERR;
 		break;
+	case IRDMA_AE_IB_REMOTE_OP_ERROR:
+		qp->flush_code = FLUSH_REM_OP_ERR;
+		break;
 	default:
 		qp->flush_code = FLUSH_FATAL_ERR;
 		break;
-- 
1.8.3.1

[PATCH for-rc 2/2] RDMA/irdma: Don't arm the CQ more than two times if no CE for this CQ

[Shiraz Saleem]

From: Tatyana Nikolova <tatyana.e.nikolova@intel.com>

Completion events (CEs) are lost if the application is allowed to arm the
CQ more than two times when no new CE for this CQ has been generated by
the HW.

Check if arming has been done for the CQ and if not, arm the CQ for any
event otherwise promote to arm the CQ for any event only when the last arm
event was solicited.

Fixes: b48c24c2d710 ("RDMA/irdma: Implement device supported verb APIs")
Signed-off-by: Tatyana Nikolova <tatyana.e.nikolova@intel.com>
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
---
 drivers/infiniband/hw/irdma/hw.c    |  2 ++
 drivers/infiniband/hw/irdma/main.h  |  1 +
 drivers/infiniband/hw/irdma/utils.c | 15 +++++++++++++++
 drivers/infiniband/hw/irdma/verbs.c | 23 ++++++++++++++++++-----
 drivers/infiniband/hw/irdma/verbs.h |  2 ++
 5 files changed, 38 insertions(+), 5 deletions(-)

diff --git a/drivers/infiniband/hw/irdma/hw.c b/drivers/infiniband/hw/irdma/hw.c
index 1bae1dc..b4c657f 100644
--- a/drivers/infiniband/hw/irdma/hw.c
+++ b/drivers/infiniband/hw/irdma/hw.c
@@ -60,6 +60,8 @@ static void irdma_iwarp_ce_handler(struct irdma_sc_cq *iwcq)
 {
 	struct irdma_cq *cq = iwcq->back_cq;
 
+	if (!cq->user_mode)
+		cq->armed = false;
 	if (cq->ibcq.comp_handler)
 		cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
 }
diff --git a/drivers/infiniband/hw/irdma/main.h b/drivers/infiniband/hw/irdma/main.h
index 91a4971..cb218ca 100644
--- a/drivers/infiniband/hw/irdma/main.h
+++ b/drivers/infiniband/hw/irdma/main.h
@@ -542,6 +542,7 @@ int irdma_ah_cqp_op(struct irdma_pci_f *rf, struct irdma_sc_ah *sc_ah, u8 cmd,
 		    void (*callback_fcn)(struct irdma_cqp_request *cqp_request),
 		    void *cb_param);
 void irdma_gsi_ud_qp_ah_cb(struct irdma_cqp_request *cqp_request);
+bool irdma_cq_empty(struct irdma_cq *iwcq);
 int irdma_inetaddr_event(struct notifier_block *notifier, unsigned long event,
 			 void *ptr);
 int irdma_inet6addr_event(struct notifier_block *notifier, unsigned long event,
diff --git a/drivers/infiniband/hw/irdma/utils.c b/drivers/infiniband/hw/irdma/utils.c
index 8b42c43..ad96ed0 100644
--- a/drivers/infiniband/hw/irdma/utils.c
+++ b/drivers/infiniband/hw/irdma/utils.c
@@ -2491,3 +2491,18 @@ void irdma_ib_qp_event(struct irdma_qp *iwqp, enum irdma_qp_event_type event)
 	ibevent.element.qp = &iwqp->ibqp;
 	iwqp->ibqp.event_handler(&ibevent, iwqp->ibqp.qp_context);
 }
+
+bool irdma_cq_empty(struct irdma_cq *iwcq)
+{
+	struct irdma_cq_uk *ukcq;
+	u64 qword3;
+	__le64 *cqe;
+	u8 polarity;
+
+	ukcq  = &iwcq->sc_cq.cq_uk;
+	cqe = IRDMA_GET_CURRENT_CQ_ELEM(ukcq);
+	get_64bit_val(cqe, 24, &qword3);
+	polarity = (u8)FIELD_GET(IRDMA_CQ_VALID, qword3);
+
+	return polarity != ukcq->polarity;
+}
diff --git a/drivers/infiniband/hw/irdma/verbs.c b/drivers/infiniband/hw/irdma/verbs.c
index 0f66e80..8cd5f92 100644
--- a/drivers/infiniband/hw/irdma/verbs.c
+++ b/drivers/infiniband/hw/irdma/verbs.c
@@ -3584,18 +3584,31 @@ static int irdma_req_notify_cq(struct ib_cq *ibcq,
 	struct irdma_cq *iwcq;
 	struct irdma_cq_uk *ukcq;
 	unsigned long flags;
-	enum irdma_cmpl_notify cq_notify = IRDMA_CQ_COMPL_EVENT;
+	enum irdma_cmpl_notify cq_notify;
+	bool promo_event = false;
+	int ret = 0;
 
+	cq_notify = notify_flags == IB_CQ_SOLICITED ?
+		    IRDMA_CQ_COMPL_SOLICITED : IRDMA_CQ_COMPL_EVENT;
 	iwcq = to_iwcq(ibcq);
 	ukcq = &iwcq->sc_cq.cq_uk;
-	if (notify_flags == IB_CQ_SOLICITED)
-		cq_notify = IRDMA_CQ_COMPL_SOLICITED;
 
 	spin_lock_irqsave(&iwcq->lock, flags);
-	irdma_uk_cq_request_notification(ukcq, cq_notify);
+	/* Only promote to arm the CQ for any event if the last arm event was solicited. */
+	if (iwcq->last_notify == IRDMA_CQ_COMPL_SOLICITED && notify_flags != IB_CQ_SOLICITED)
+		promo_event = true;
+
+	if (!iwcq->armed || promo_event) {
+		iwcq->armed = true;
+		iwcq->last_notify = cq_notify;
+		irdma_uk_cq_request_notification(ukcq, cq_notify);
+	}
+
+	if ((notify_flags & IB_CQ_REPORT_MISSED_EVENTS) && !irdma_cq_empty(iwcq))
+		ret = 1;
 	spin_unlock_irqrestore(&iwcq->lock, flags);
 
-	return 0;
+	return ret;
 }
 
 static int irdma_roce_port_immutable(struct ib_device *ibdev, u32 port_num,
diff --git a/drivers/infiniband/hw/irdma/verbs.h b/drivers/infiniband/hw/irdma/verbs.h
index 5c244cd..d0fdef8 100644
--- a/drivers/infiniband/hw/irdma/verbs.h
+++ b/drivers/infiniband/hw/irdma/verbs.h
@@ -110,6 +110,8 @@ struct irdma_cq {
 	u16 cq_size;
 	u16 cq_num;
 	bool user_mode;
+	bool armed;
+	enum irdma_cmpl_notify last_notify;
 	u32 polled_cmpls;
 	u32 cq_mem_size;
 	struct irdma_dma_mem kmem;
-- 
1.8.3.1

Re: [PATCH for-rc 1/2] RDMA/irdma: Report correct WC errors

[Jason Gunthorpe]

On Wed, Dec 01, 2021 at 05:15:08PM -0600, Shiraz Saleem wrote:
> Return IBV_WC_REM_OP_ERR for responder QP errors instead of
> IBV_WC_REM_ACCESS_ERR.
> 
> Return IBV_WC_LOC_QP_OP_ERR for errors detected on the SQ with bad opcodes
> 
> Fixes: 44d9e52977a1 ("RDMA/irdma: Implement device initialization definitions")
> Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
> ---
>  drivers/infiniband/hw/irdma/hw.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)

This and the 2/2 applied to for-rc, thanks

Jason