1 files changed, 239 insertions, 121 deletions

diff --git a/drivers/infiniband/hw/cxgb4/cq.c b/drivers/infiniband/hw/cxgb4/cq.c
index 8d8f8add6fc..c04292c950f 100644
--- a/drivers/infiniband/hw/cxgb4/cq.c
+++ b/drivers/infiniband/hw/cxgb4/cq.c
@@ -134,7 +134,8 @@ static int create_cq(struct c4iw_rdev *rdev, struct t4_cq *cq,
 			V_FW_RI_RES_WR_IQANUS(0) |
 			V_FW_RI_RES_WR_IQANUD(1) |
 			F_FW_RI_RES_WR_IQANDST |
-			V_FW_RI_RES_WR_IQANDSTINDEX(*rdev->lldi.rxq_ids));
+			V_FW_RI_RES_WR_IQANDSTINDEX(
+				rdev->lldi.ciq_ids[cq->vector]));
 	res->u.cq.iqdroprss_to_iqesize = cpu_to_be16(
 			F_FW_RI_RES_WR_IQDROPRSS |
 			V_FW_RI_RES_WR_IQPCIECH(2) |
@@ -185,7 +186,7 @@ static void insert_recv_cqe(struct t4_wq *wq, struct t4_cq *cq)
 				 V_CQE_OPCODE(FW_RI_SEND) |
 				 V_CQE_TYPE(0) |
 				 V_CQE_SWCQE(1) |
-				 V_CQE_QPID(wq->rq.qid));
+				 V_CQE_QPID(wq->sq.qid));
 	cqe.bits_type_ts = cpu_to_be64(V_CQE_GENBIT((u64)cq->gen));
 	cq->sw_queue[cq->sw_pidx] = cqe;
 	t4_swcq_produce(cq);
@@ -225,43 +226,186 @@ static void insert_sq_cqe(struct t4_wq *wq, struct t4_cq *cq,
 	t4_swcq_produce(cq);
 }
 
-int c4iw_flush_sq(struct t4_wq *wq, struct t4_cq *cq, int count)
+static void advance_oldest_read(struct t4_wq *wq);
+
+int c4iw_flush_sq(struct c4iw_qp *qhp)
 {
 	int flushed = 0;
-	struct t4_swsqe *swsqe = &wq->sq.sw_sq[wq->sq.cidx + count];
-	int in_use = wq->sq.in_use - count;
+	struct t4_wq *wq = &qhp->wq;
+	struct c4iw_cq *chp = to_c4iw_cq(qhp->ibqp.send_cq);
+	struct t4_cq *cq = &chp->cq;
+	int idx;
+	struct t4_swsqe *swsqe;
 
-	BUG_ON(in_use < 0);
-	while (in_use--) {
-		swsqe->signaled = 0;
+	if (wq->sq.flush_cidx == -1)
+		wq->sq.flush_cidx = wq->sq.cidx;
+	idx = wq->sq.flush_cidx;
+	BUG_ON(idx >= wq->sq.size);
+	while (idx != wq->sq.pidx) {
+		swsqe = &wq->sq.sw_sq[idx];
+		BUG_ON(swsqe->flushed);
+		swsqe->flushed = 1;
 		insert_sq_cqe(wq, cq, swsqe);
-		swsqe++;
-		if (swsqe == (wq->sq.sw_sq + wq->sq.size))
-			swsqe = wq->sq.sw_sq;
+		if (wq->sq.oldest_read == swsqe) {
+			BUG_ON(swsqe->opcode != FW_RI_READ_REQ);
+			advance_oldest_read(wq);
+		}
 		flushed++;
+		if (++idx == wq->sq.size)
+			idx = 0;
 	}
+	wq->sq.flush_cidx += flushed;
+	if (wq->sq.flush_cidx >= wq->sq.size)
+		wq->sq.flush_cidx -= wq->sq.size;
 	return flushed;
 }
 
+static void flush_completed_wrs(struct t4_wq *wq, struct t4_cq *cq)
+{
+	struct t4_swsqe *swsqe;
+	int cidx;
+
+	if (wq->sq.flush_cidx == -1)
+		wq->sq.flush_cidx = wq->sq.cidx;
+	cidx = wq->sq.flush_cidx;
+	BUG_ON(cidx > wq->sq.size);
+
+	while (cidx != wq->sq.pidx) {
+		swsqe = &wq->sq.sw_sq[cidx];
+		if (!swsqe->signaled) {
+			if (++cidx == wq->sq.size)
+				cidx = 0;
+		} else if (swsqe->complete) {
+
+			BUG_ON(swsqe->flushed);
+
+			/*
+			 * Insert this completed cqe into the swcq.
+			 */
+			PDBG("%s moving cqe into swcq sq idx %u cq idx %u\n",
+					__func__, cidx, cq->sw_pidx);
+			swsqe->cqe.header |= htonl(V_CQE_SWCQE(1));
+			cq->sw_queue[cq->sw_pidx] = swsqe->cqe;
+			t4_swcq_produce(cq);
+			swsqe->flushed = 1;
+			if (++cidx == wq->sq.size)
+				cidx = 0;
+			wq->sq.flush_cidx = cidx;
+		} else
+			break;
+	}
+}
+
+static void create_read_req_cqe(struct t4_wq *wq, struct t4_cqe *hw_cqe,
+		struct t4_cqe *read_cqe)
+{
+	read_cqe->u.scqe.cidx = wq->sq.oldest_read->idx;
+	read_cqe->len = htonl(wq->sq.oldest_read->read_len);
+	read_cqe->header = htonl(V_CQE_QPID(CQE_QPID(hw_cqe)) |
+			V_CQE_SWCQE(SW_CQE(hw_cqe)) |
+			V_CQE_OPCODE(FW_RI_READ_REQ) |
+			V_CQE_TYPE(1));
+	read_cqe->bits_type_ts = hw_cqe->bits_type_ts;
+}
+
+static void advance_oldest_read(struct t4_wq *wq)
+{
+
+	u32 rptr = wq->sq.oldest_read - wq->sq.sw_sq + 1;
+
+	if (rptr == wq->sq.size)
+		rptr = 0;
+	while (rptr != wq->sq.pidx) {
+		wq->sq.oldest_read = &wq->sq.sw_sq[rptr];
+
+		if (wq->sq.oldest_read->opcode == FW_RI_READ_REQ)
+			return;
+		if (++rptr == wq->sq.size)
+			rptr = 0;
+	}
+	wq->sq.oldest_read = NULL;
+}
+
 /*
  * Move all CQEs from the HWCQ into the SWCQ.
+ * Deal with out-of-order and/or completions that complete
+ * prior unsignalled WRs.
  */
-void c4iw_flush_hw_cq(struct t4_cq *cq)
+void c4iw_flush_hw_cq(struct c4iw_cq *chp)
 {
-	struct t4_cqe *cqe = NULL, *swcqe;
+	struct t4_cqe *hw_cqe, *swcqe, read_cqe;
+	struct c4iw_qp *qhp;
+	struct t4_swsqe *swsqe;
 	int ret;
 
-	PDBG("%s cq %p cqid 0x%x\n", __func__, cq, cq->cqid);
-	ret = t4_next_hw_cqe(cq, &cqe);
+	PDBG("%s  cqid 0x%x\n", __func__, chp->cq.cqid);
+	ret = t4_next_hw_cqe(&chp->cq, &hw_cqe);
+
+	/*
+	 * This logic is similar to poll_cq(), but not quite the same
+	 * unfortunately.  Need to move pertinent HW CQEs to the SW CQ but
+	 * also do any translation magic that poll_cq() normally does.
+	 */
 	while (!ret) {
-		PDBG("%s flushing hwcq cidx 0x%x swcq pidx 0x%x\n",
-		     __func__, cq->cidx, cq->sw_pidx);
-		swcqe = &cq->sw_queue[cq->sw_pidx];
-		*swcqe = *cqe;
-		swcqe->header |= cpu_to_be32(V_CQE_SWCQE(1));
-		t4_swcq_produce(cq);
-		t4_hwcq_consume(cq);
-		ret = t4_next_hw_cqe(cq, &cqe);
+		qhp = get_qhp(chp->rhp, CQE_QPID(hw_cqe));
+
+		/*
+		 * drop CQEs with no associated QP
+		 */
+		if (qhp == NULL)
+			goto next_cqe;
+
+		if (CQE_OPCODE(hw_cqe) == FW_RI_TERMINATE)
+			goto next_cqe;
+
+		if (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)
+				goto next_cqe;
+
+			/* drop peer2peer RTR reads.
+			 */
+			if (CQE_WRID_STAG(hw_cqe) == 1)
+				goto next_cqe;
+
+			/*
+			 * Eat completions for unsignaled read WRs.
+			 */
+			if (!qhp->wq.sq.oldest_read->signaled) {
+				advance_oldest_read(&qhp->wq);
+				goto next_cqe;
+			}
+
+			/*
+			 * Don't write to the HWCQ, create a new read req CQE
+			 * in local memory and move it into the swcq.
+			 */
+			create_read_req_cqe(&qhp->wq, hw_cqe, &read_cqe);
+			hw_cqe = &read_cqe;
+			advance_oldest_read(&qhp->wq);
+		}
+
+		/* if its a SQ completion, then do the magic to move all the
+		 * unsignaled and now in-order completions into the swcq.
+		 */
+		if (SQ_TYPE(hw_cqe)) {
+			swsqe = &qhp->wq.sq.sw_sq[CQE_WRID_SQ_IDX(hw_cqe)];
+			swsqe->cqe = *hw_cqe;
+			swsqe->complete = 1;
+			flush_completed_wrs(&qhp->wq, &chp->cq);
+		} else {
+			swcqe = &chp->cq.sw_queue[chp->cq.sw_pidx];
+			*swcqe = *hw_cqe;
+			swcqe->header |= cpu_to_be32(V_CQE_SWCQE(1));
+			t4_swcq_produce(&chp->cq);
+		}
+next_cqe:
+		t4_hwcq_consume(&chp->cq);
+		ret = t4_next_hw_cqe(&chp->cq, &hw_cqe);
 	}
 }
 
@@ -281,25 +425,6 @@ static int cqe_completes_wr(struct t4_cqe *cqe, struct t4_wq *wq)
 	return 1;
 }
 
-void c4iw_count_scqes(struct t4_cq *cq, struct t4_wq *wq, int *count)
-{
-	struct t4_cqe *cqe;
-	u32 ptr;
-
-	*count = 0;
-	ptr = cq->sw_cidx;
-	while (ptr != cq->sw_pidx) {
-		cqe = &cq->sw_queue[ptr];
-		if ((SQ_TYPE(cqe) || ((CQE_OPCODE(cqe) == FW_RI_READ_RESP) &&
-				      wq->sq.oldest_read)) &&
-		    (CQE_QPID(cqe) == wq->sq.qid))
-			(*count)++;
-		if (++ptr == cq->size)
-			ptr = 0;
-	}
-	PDBG("%s cq %p count %d\n", __func__, cq, *count);
-}
-
 void c4iw_count_rcqes(struct t4_cq *cq, struct t4_wq *wq, int *count)
 {
 	struct t4_cqe *cqe;
@@ -311,7 +436,7 @@ void c4iw_count_rcqes(struct t4_cq *cq, struct t4_wq *wq, int *count)
 	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))
+		    (CQE_QPID(cqe) == wq->sq.qid) && cqe_completes_wr(cqe, wq))
 			(*count)++;
 		if (++ptr == cq->size)
 			ptr = 0;
@@ -319,70 +444,6 @@ void c4iw_count_rcqes(struct t4_cq *cq, struct t4_wq *wq, int *count)
 	PDBG("%s cq %p count %d\n", __func__, cq, *count);
 }
 
-static void flush_completed_wrs(struct t4_wq *wq, struct t4_cq *cq)
-{
-	struct t4_swsqe *swsqe;
-	u16 ptr = wq->sq.cidx;
-	int count = wq->sq.in_use;
-	int unsignaled = 0;
-
-	swsqe = &wq->sq.sw_sq[ptr];
-	while (count--)
-		if (!swsqe->signaled) {
-			if (++ptr == wq->sq.size)
-				ptr = 0;
-			swsqe = &wq->sq.sw_sq[ptr];
-			unsignaled++;
-		} else if (swsqe->complete) {
-
-			/*
-			 * Insert this completed cqe into the swcq.
-			 */
-			PDBG("%s moving cqe into swcq sq idx %u cq idx %u\n",
-			     __func__, ptr, cq->sw_pidx);
-			swsqe->cqe.header |= htonl(V_CQE_SWCQE(1));
-			cq->sw_queue[cq->sw_pidx] = swsqe->cqe;
-			t4_swcq_produce(cq);
-			swsqe->signaled = 0;
-			wq->sq.in_use -= unsignaled;
-			break;
-		} else
-			break;
-}
-
-static void create_read_req_cqe(struct t4_wq *wq, struct t4_cqe *hw_cqe,
-				struct t4_cqe *read_cqe)
-{
-	read_cqe->u.scqe.cidx = wq->sq.oldest_read->idx;
-	read_cqe->len = cpu_to_be32(wq->sq.oldest_read->read_len);
-	read_cqe->header = htonl(V_CQE_QPID(CQE_QPID(hw_cqe)) |
-				 V_CQE_SWCQE(SW_CQE(hw_cqe)) |
-				 V_CQE_OPCODE(FW_RI_READ_REQ) |
-				 V_CQE_TYPE(1));
-	read_cqe->bits_type_ts = hw_cqe->bits_type_ts;
-}
-
-/*
- * Return a ptr to the next read wr in the SWSQ or NULL.
- */
-static void advance_oldest_read(struct t4_wq *wq)
-{
-
-	u32 rptr = wq->sq.oldest_read - wq->sq.sw_sq + 1;
-
-	if (rptr == wq->sq.size)
-		rptr = 0;
-	while (rptr != wq->sq.pidx) {
-		wq->sq.oldest_read = &wq->sq.sw_sq[rptr];
-
-		if (wq->sq.oldest_read->opcode == FW_RI_READ_REQ)
-			return;
-		if (++rptr == wq->sq.size)
-			rptr = 0;
-	}
-	wq->sq.oldest_read = NULL;
-}
-
 /*
  * poll_cq
  *
@@ -427,6 +488,22 @@ static int poll_cq(struct t4_wq *wq, struct t4_cq *cq, struct t4_cqe *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;
+	}
+
+	/*
 	 * Gotta tweak READ completions:
 	 *	1) the cqe doesn't contain the sq_wptr from the wr.
 	 *	2) opcode not reflected from the wr.
@@ -435,12 +512,22 @@ static int poll_cq(struct t4_wq *wq, struct t4_cq *cq, struct t4_cqe *cqe,
 	 */
 	if (RQ_TYPE(hw_cqe) && (CQE_OPCODE(hw_cqe) == FW_RI_READ_RESP)) {
 
-		/*
-		 * If this is an unsolicited read response, then the read
+		/* 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);
+			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 (!wq->sq.oldest_read) {
+		if (CQE_WRID_STAG(hw_cqe) == 1) {
 			if (CQE_STATUS(hw_cqe))
 				t4_set_wq_in_error(wq);
 			ret = -EAGAIN;
@@ -448,6 +535,15 @@ static int poll_cq(struct t4_wq *wq, struct t4_cq *cq, struct t4_cqe *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.
 		 */
@@ -457,14 +553,8 @@ static int poll_cq(struct t4_wq *wq, struct t4_cq *cq, struct t4_cqe *cqe,
 	}
 
 	if (CQE_STATUS(hw_cqe) || t4_wq_in_error(wq)) {
-		*cqe_flushed = t4_wq_in_error(wq);
+		*cqe_flushed = (CQE_STATUS(hw_cqe) == T4_ERR_SWFLUSH);
 		t4_set_wq_in_error(wq);
-		goto proc_cqe;
-	}
-
-	if (CQE_OPCODE(hw_cqe) == FW_RI_TERMINATE) {
-		ret = -EAGAIN;
-		goto skip_cqe;
 	}
 
 	/*
@@ -523,7 +613,24 @@ proc_cqe:
 	 * completion.
 	 */
 	if (SQ_TYPE(hw_cqe)) {
-		wq->sq.cidx = CQE_WRID_SQ_IDX(hw_cqe);
+		int idx = CQE_WRID_SQ_IDX(hw_cqe);
+		BUG_ON(idx >= wq->sq.size);
+
+		/*
+		* 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;
+		BUG_ON(wq->sq.in_use <= 0 && wq->sq.in_use >= wq->sq.size);
+
+		wq->sq.cidx = (uint16_t)idx;
 		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);
@@ -532,6 +639,7 @@ proc_cqe:
 		*cookie = wq->rq.sw_rq[wq->rq.cidx].wr_id;
 		BUG_ON(t4_rq_empty(wq));
 		t4_rq_consume(wq);
+		goto skip_cqe;
 	}
 
 flush_wq:
@@ -565,7 +673,7 @@ skip_cqe:
 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_cqe uninitialized_var(cqe), *rd_cqe;
 	struct t4_wq *wq;
 	u32 credit = 0;
 	u8 cqe_flushed;
@@ -763,6 +871,9 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev, int entries,
 
 	rhp = to_c4iw_dev(ibdev);
 
+	if (vector >= rhp->rdev.lldi.nciq)
+		return ERR_PTR(-EINVAL);
+
 	chp = kzalloc(sizeof(*chp), GFP_KERNEL);
 	if (!chp)
 		return ERR_PTR(-ENOMEM);
@@ -784,7 +895,7 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev, int entries,
 	/*
 	 * Make actual HW queue 2x to avoid cdix_inc overflows.
 	 */
-	hwentries = entries * 2;
+	hwentries = min(entries * 2, T4_MAX_IQ_SIZE);
 
 	/*
 	 * Make HW queue at least 64 entries so GTS updates aren't too
@@ -801,9 +912,14 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev, int entries,
 	if (ucontext) {
 		memsize = roundup(memsize, PAGE_SIZE);
 		hwentries = memsize / sizeof *chp->cq.queue;
+		while (hwentries > T4_MAX_IQ_SIZE) {
+			memsize -= PAGE_SIZE;
+			hwentries = memsize / sizeof *chp->cq.queue;
+		}
 	}
 	chp->cq.size = hwentries;
 	chp->cq.memsize = memsize;
+	chp->cq.vector = vector;
 
 	ret = create_cq(&rhp->rdev, &chp->cq,
 			ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
@@ -814,6 +930,7 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev, int entries,
 	chp->cq.size--;				/* status page */
 	chp->ibcq.cqe = entries - 2;
 	spin_lock_init(&chp->lock);
+	spin_lock_init(&chp->comp_handler_lock);
 	atomic_set(&chp->refcnt, 1);
 	init_waitqueue_head(&chp->wait);
 	ret = insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid);
@@ -838,7 +955,8 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev, int entries,
 		uresp.gts_key = ucontext->key;
 		ucontext->key += PAGE_SIZE;
 		spin_unlock(&ucontext->mmap_lock);
-		ret = ib_copy_to_udata(udata, &uresp, sizeof uresp);
+		ret = ib_copy_to_udata(udata, &uresp,
+				       sizeof(uresp) - sizeof(uresp.reserved));
 		if (ret)
 			goto err5;