1 files changed, 209 insertions, 60 deletions

diff --git a/drivers/infiniband/hw/cxgb4/cq.c b/drivers/infiniband/hw/cxgb4/cq.c
index 2be2e1ac1b5f..6d3042794094 100644
--- a/drivers/infiniband/hw/cxgb4/cq.c
+++ b/drivers/infiniband/hw/cxgb4/cq.c
@@ -77,6 +77,10 @@ static int create_cq(struct c4iw_rdev *rdev, struct t4_cq *cq,
 	int user = (uctx != &rdev->uctx);
 	int ret;
 	struct sk_buff *skb;
+	struct c4iw_ucontext *ucontext = NULL;
+
+	if (user)
+		ucontext = container_of(uctx, struct c4iw_ucontext, uctx);
 
 	cq->cqid = c4iw_get_cqid(rdev, uctx);
 	if (!cq->cqid) {
@@ -100,6 +104,16 @@ static int create_cq(struct c4iw_rdev *rdev, struct t4_cq *cq,
 	dma_unmap_addr_set(cq, mapping, cq->dma_addr);
 	memset(cq->queue, 0, cq->memsize);
 
+	if (user && ucontext->is_32b_cqe) {
+		cq->qp_errp = &((struct t4_status_page *)
+		((u8 *)cq->queue + (cq->size - 1) *
+		 (sizeof(*cq->queue) / 2)))->qp_err;
+	} else {
+		cq->qp_errp = &((struct t4_status_page *)
+		((u8 *)cq->queue + (cq->size - 1) *
+		 sizeof(*cq->queue)))->qp_err;
+	}
+
 	/* build fw_ri_res_wr */
 	wr_len = sizeof *res_wr + sizeof *res;
 
@@ -132,7 +146,9 @@ static int create_cq(struct c4iw_rdev *rdev, struct t4_cq *cq,
 			FW_RI_RES_WR_IQPCIECH_V(2) |
 			FW_RI_RES_WR_IQINTCNTTHRESH_V(0) |
 			FW_RI_RES_WR_IQO_F |
-			FW_RI_RES_WR_IQESIZE_V(1));
+			((user && ucontext->is_32b_cqe) ?
+			 FW_RI_RES_WR_IQESIZE_V(1) :
+			 FW_RI_RES_WR_IQESIZE_V(2)));
 	res->u.cq.iqsize = cpu_to_be16(cq->size);
 	res->u.cq.iqaddr = cpu_to_be64(cq->dma_addr);
 
@@ -166,7 +182,7 @@ err1:
 	return ret;
 }
 
-static void insert_recv_cqe(struct t4_wq *wq, struct t4_cq *cq)
+static void insert_recv_cqe(struct t4_wq *wq, struct t4_cq *cq, u32 srqidx)
 {
 	struct t4_cqe cqe;
 
@@ -179,6 +195,8 @@ static void insert_recv_cqe(struct t4_wq *wq, struct t4_cq *cq)
 				 CQE_SWCQE_V(1) |
 				 CQE_QPID_V(wq->sq.qid));
 	cqe.bits_type_ts = cpu_to_be64(CQE_GENBIT_V((u64)cq->gen));
+	if (srqidx)
+		cqe.u.srcqe.abs_rqe_idx = cpu_to_be32(srqidx);
 	cq->sw_queue[cq->sw_pidx] = cqe;
 	t4_swcq_produce(cq);
 }
@@ -191,7 +209,7 @@ int c4iw_flush_rq(struct t4_wq *wq, struct t4_cq *cq, int count)
 	pr_debug("wq %p cq %p rq.in_use %u skip count %u\n",
 		 wq, cq, wq->rq.in_use, count);
 	while (in_use--) {
-		insert_recv_cqe(wq, cq);
+		insert_recv_cqe(wq, cq, 0);
 		flushed++;
 	}
 	return flushed;
@@ -442,6 +460,72 @@ void c4iw_count_rcqes(struct t4_cq *cq, struct t4_wq *wq, int *count)
 	pr_debug("cq %p count %d\n", cq, *count);
 }
 
+static void post_pending_srq_wrs(struct t4_srq *srq)
+{
+	struct t4_srq_pending_wr *pwr;
+	u16 idx = 0;
+
+	while (srq->pending_in_use) {
+		pwr = &srq->pending_wrs[srq->pending_cidx];
+		srq->sw_rq[srq->pidx].wr_id = pwr->wr_id;
+		srq->sw_rq[srq->pidx].valid = 1;
+
+		pr_debug("%s posting pending cidx %u pidx %u wq_pidx %u in_use %u rq_size %u wr_id %llx\n",
+			 __func__,
+			 srq->cidx, srq->pidx, srq->wq_pidx,
+			 srq->in_use, srq->size,
+			 (unsigned long long)pwr->wr_id);
+
+		c4iw_copy_wr_to_srq(srq, &pwr->wqe, pwr->len16);
+		t4_srq_consume_pending_wr(srq);
+		t4_srq_produce(srq, pwr->len16);
+		idx += DIV_ROUND_UP(pwr->len16 * 16, T4_EQ_ENTRY_SIZE);
+	}
+
+	if (idx) {
+		t4_ring_srq_db(srq, idx, pwr->len16, &pwr->wqe);
+		srq->queue[srq->size].status.host_wq_pidx =
+			srq->wq_pidx;
+	}
+}
+
+static u64 reap_srq_cqe(struct t4_cqe *hw_cqe, struct t4_srq *srq)
+{
+	int rel_idx = CQE_ABS_RQE_IDX(hw_cqe) - srq->rqt_abs_idx;
+	u64 wr_id;
+
+	srq->sw_rq[rel_idx].valid = 0;
+	wr_id = srq->sw_rq[rel_idx].wr_id;
+
+	if (rel_idx == srq->cidx) {
+		pr_debug("%s in order cqe rel_idx %u cidx %u pidx %u wq_pidx %u in_use %u rq_size %u wr_id %llx\n",
+			 __func__, rel_idx, srq->cidx, srq->pidx,
+			 srq->wq_pidx, srq->in_use, srq->size,
+			 (unsigned long long)srq->sw_rq[rel_idx].wr_id);
+		t4_srq_consume(srq);
+		while (srq->ooo_count && !srq->sw_rq[srq->cidx].valid) {
+			pr_debug("%s eat ooo cidx %u pidx %u wq_pidx %u in_use %u rq_size %u ooo_count %u wr_id %llx\n",
+				 __func__, srq->cidx, srq->pidx,
+				 srq->wq_pidx, srq->in_use,
+				 srq->size, srq->ooo_count,
+				 (unsigned long long)
+				 srq->sw_rq[srq->cidx].wr_id);
+			t4_srq_consume_ooo(srq);
+		}
+		if (srq->ooo_count == 0 && srq->pending_in_use)
+			post_pending_srq_wrs(srq);
+	} else {
+		pr_debug("%s ooo cqe rel_idx %u cidx %u pidx %u wq_pidx %u in_use %u rq_size %u ooo_count %u wr_id %llx\n",
+			 __func__, rel_idx, srq->cidx,
+			 srq->pidx, srq->wq_pidx,
+			 srq->in_use, srq->size,
+			 srq->ooo_count,
+			 (unsigned long long)srq->sw_rq[rel_idx].wr_id);
+		t4_srq_produce_ooo(srq);
+	}
+	return wr_id;
+}
+
 /*
  * poll_cq
  *
@@ -459,7 +543,8 @@ void c4iw_count_rcqes(struct t4_cq *cq, struct t4_wq *wq, int *count)
  *    -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)
+		   u8 *cqe_flushed, u64 *cookie, u32 *credit,
+		   struct t4_srq *srq)
 {
 	int ret = 0;
 	struct t4_cqe *hw_cqe, read_cqe;
@@ -524,7 +609,7 @@ static int poll_cq(struct t4_wq *wq, struct t4_cq *cq, struct t4_cqe *cqe,
 		 */
 		if (CQE_TYPE(hw_cqe) == 1) {
 			if (CQE_STATUS(hw_cqe))
-				t4_set_wq_in_error(wq);
+				t4_set_wq_in_error(wq, 0);
 			ret = -EAGAIN;
 			goto skip_cqe;
 		}
@@ -535,7 +620,7 @@ static int poll_cq(struct t4_wq *wq, struct t4_cq *cq, struct t4_cqe *cqe,
 		 */
 		if (CQE_WRID_STAG(hw_cqe) == 1) {
 			if (CQE_STATUS(hw_cqe))
-				t4_set_wq_in_error(wq);
+				t4_set_wq_in_error(wq, 0);
 			ret = -EAGAIN;
 			goto skip_cqe;
 		}
@@ -560,7 +645,7 @@ 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 = (CQE_STATUS(hw_cqe) == T4_ERR_SWFLUSH);
-		t4_set_wq_in_error(wq);
+		t4_set_wq_in_error(wq, 0);
 	}
 
 	/*
@@ -574,15 +659,9 @@ static int poll_cq(struct t4_wq *wq, struct t4_cq *cq, struct t4_cqe *cqe,
 		 * 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_STATUS(hw_cqe) &&
 			     CQE_WRID_MSN(hw_cqe) != wq->rq.msn)) {
-			t4_set_wq_in_error(wq);
+			t4_set_wq_in_error(wq, 0);
 			hw_cqe->header |= cpu_to_be32(CQE_STATUS_V(T4_ERR_MSN));
 		}
 		goto proc_cqe;
@@ -641,11 +720,16 @@ proc_cqe:
 			c4iw_log_wr_stats(wq, hw_cqe);
 		t4_sq_consume(wq);
 	} else {
-		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);
+		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;
 	}
 
@@ -668,46 +752,33 @@ skip_cqe:
 	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)
+static int __c4iw_poll_cq_one(struct c4iw_cq *chp, struct c4iw_qp *qhp,
+			      struct ib_wc *wc, struct c4iw_srq *srq)
 {
-	struct c4iw_qp *qhp = NULL;
-	struct t4_cqe uninitialized_var(cqe), *rd_cqe;
-	struct t4_wq *wq;
+	struct t4_cqe uninitialized_var(cqe);
+	struct t4_wq *wq = qhp ? &qhp->wq : NULL;
 	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);
+	ret = poll_cq(wq, &(chp->cq), &cqe, &cqe_flushed, &cookie, &credit,
+		      srq ? &srq->wq : NULL);
 	if (ret)
 		goto out;
 
 	wc->wr_id = cookie;
-	wc->qp = &qhp->ibqp;
+	wc->qp = qhp ? &qhp->ibqp : NULL;
 	wc->vendor_err = CQE_STATUS(&cqe);
 	wc->wc_flags = 0;
 
+	/*
+	 * Simulate a SRQ_LIMIT_REACHED HW notification if required.
+	 */
+	if (srq && !(srq->flags & T4_SRQ_LIMIT_SUPPORT) && srq->armed &&
+	    srq->wq.in_use < srq->srq_limit)
+		c4iw_dispatch_srq_limit_reached_event(srq);
+
 	pr_debug("qpid 0x%x type %d opcode %d status 0x%x len %u wrid hi 0x%x lo 0x%x cookie 0x%llx\n",
 		 CQE_QPID(&cqe),
 		 CQE_TYPE(&cqe), CQE_OPCODE(&cqe),
@@ -720,15 +791,32 @@ static int c4iw_poll_cq_one(struct c4iw_cq *chp, struct ib_wc *wc)
 			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) {
+
+		switch (CQE_OPCODE(&cqe)) {
+		case FW_RI_SEND:
+			wc->opcode = IB_WC_RECV;
+			break;
+		case FW_RI_SEND_WITH_INV:
+		case FW_RI_SEND_WITH_SE_INV:
+			wc->opcode = IB_WC_RECV;
 			wc->ex.invalidate_rkey = CQE_WRID_STAG(&cqe);
 			wc->wc_flags |= IB_WC_WITH_INVALIDATE;
 			c4iw_invalidate_mr(qhp->rhp, wc->ex.invalidate_rkey);
+			break;
+		case FW_RI_WRITE_IMMEDIATE:
+			wc->opcode = IB_WC_RECV_RDMA_WITH_IMM;
+			wc->ex.imm_data = CQE_IMM_DATA(&cqe);
+			wc->wc_flags |= IB_WC_WITH_IMM;
+			break;
+		default:
+			pr_err("Unexpected opcode %d in the CQE received for QPID=0x%0x\n",
+			       CQE_OPCODE(&cqe), CQE_QPID(&cqe));
+			ret = -EINVAL;
+			goto out;
 		}
 	} else {
 		switch (CQE_OPCODE(&cqe)) {
+		case FW_RI_WRITE_IMMEDIATE:
 		case FW_RI_RDMA_WRITE:
 			wc->opcode = IB_WC_RDMA_WRITE;
 			break;
@@ -819,8 +907,43 @@ static int c4iw_poll_cq_one(struct c4iw_cq *chp, struct ib_wc *wc)
 		}
 	}
 out:
-	if (wq)
+	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_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;
 }
 
@@ -876,6 +999,7 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev,
 	int vector = attr->comp_vector;
 	struct c4iw_dev *rhp;
 	struct c4iw_cq *chp;
+	struct c4iw_create_cq ucmd;
 	struct c4iw_create_cq_resp uresp;
 	struct c4iw_ucontext *ucontext = NULL;
 	int ret, wr_len;
@@ -891,9 +1015,16 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev,
 	if (vector >= rhp->rdev.lldi.nciq)
 		return ERR_PTR(-EINVAL);
 
+	if (ib_context) {
+		ucontext = to_c4iw_ucontext(ib_context);
+		if (udata->inlen < sizeof(ucmd))
+			ucontext->is_32b_cqe = 1;
+	}
+
 	chp = kzalloc(sizeof(*chp), GFP_KERNEL);
 	if (!chp)
 		return ERR_PTR(-ENOMEM);
+
 	chp->wr_waitp = c4iw_alloc_wr_wait(GFP_KERNEL);
 	if (!chp->wr_waitp) {
 		ret = -ENOMEM;
@@ -908,9 +1039,6 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev,
 		goto err_free_wr_wait;
 	}
 
-	if (ib_context)
-		ucontext = to_c4iw_ucontext(ib_context);
-
 	/* account for the status page. */
 	entries++;
 
@@ -934,13 +1062,15 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev,
 	if (hwentries < 64)
 		hwentries = 64;
 
-	memsize = hwentries * sizeof *chp->cq.queue;
+	memsize = hwentries * ((ucontext && ucontext->is_32b_cqe) ?
+			(sizeof(*chp->cq.queue) / 2) : sizeof(*chp->cq.queue));
 
 	/*
 	 * memsize must be a multiple of the page size if its a user cq.
 	 */
 	if (ucontext)
 		memsize = roundup(memsize, PAGE_SIZE);
+
 	chp->cq.size = hwentries;
 	chp->cq.memsize = memsize;
 	chp->cq.vector = vector;
@@ -971,6 +1101,7 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev,
 		if (!mm2)
 			goto err_free_mm;
 
+		memset(&uresp, 0, sizeof(uresp));
 		uresp.qid_mask = rhp->rdev.cqmask;
 		uresp.cqid = chp->cq.cqid;
 		uresp.size = chp->cq.size;
@@ -980,9 +1111,16 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev,
 		ucontext->key += PAGE_SIZE;
 		uresp.gts_key = ucontext->key;
 		ucontext->key += PAGE_SIZE;
+		/* communicate to the userspace that
+		 * kernel driver supports 64B CQE
+		 */
+		uresp.flags |= C4IW_64B_CQE;
+
 		spin_unlock(&ucontext->mmap_lock);
 		ret = ib_copy_to_udata(udata, &uresp,
-				       sizeof(uresp) - sizeof(uresp.reserved));
+				       ucontext->is_32b_cqe ?
+				       sizeof(uresp) - sizeof(uresp.flags) :
+				       sizeof(uresp));
 		if (ret)
 			goto err_free_mm2;
 
@@ -1019,11 +1157,6 @@ err_free_chp:
 	return ERR_PTR(ret);
 }
 
-int c4iw_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
-{
-	return -ENOSYS;
-}
-
 int c4iw_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
 {
 	struct c4iw_cq *chp;
@@ -1039,3 +1172,19 @@ int c4iw_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
 	spin_unlock_irqrestore(&chp->lock, flag);
 	return ret;
 }
+
+void c4iw_flush_srqidx(struct c4iw_qp *qhp, u32 srqidx)
+{
+	struct c4iw_cq *rchp = to_c4iw_cq(qhp->ibqp.recv_cq);
+	unsigned long flag;
+
+	/* locking heirarchy: cq lock first, then qp lock. */
+	spin_lock_irqsave(&rchp->lock, flag);
+	spin_lock(&qhp->lock);
+
+	/* create a SRQ RECV CQE for srqidx */
+	insert_recv_cqe(&qhp->wq, &rchp->cq, srqidx);
+
+	spin_unlock(&qhp->lock);
+	spin_unlock_irqrestore(&rchp->lock, flag);
+}