rdma-core/0053-libhns-Support-reporting-wc-as-software-mode.patch

From 456072a07831ca19f81b591e3c259932ece8dcae Mon Sep 17 00:00:00 2001
From: Chengchang Tang <tangchengchang@huawei.com>
Date: Tue, 26 Sep 2023 19:19:06 +0800
Subject: [PATCH 1/5] libhns: Support reporting wc as software mode

driver inclusion
category: feature
bugzilla: https://gitee.com/src-openeuler/rdma-core/issues/I83BP0

----------------------------------------------------------

When HW is in resetting stage, we could not poll back all the expected
work completions as the HW won't generate cqe anymore.

This patch allows driver to compose the expected wc instead of the HW
during resetting stage. Once the hardware finished resetting, we can
poll cq from hardware again.

Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
---
 providers/hns/hns_roce_u.h       |  12 ++
 providers/hns/hns_roce_u_hw_v2.c | 216 +++++++++++++++++++++++++++++--
 providers/hns/hns_roce_u_hw_v2.h |   2 +
 providers/hns/hns_roce_u_verbs.c |  91 +++++++++++++
 4 files changed, 309 insertions(+), 12 deletions(-)

diff --git a/providers/hns/hns_roce_u.h b/providers/hns/hns_roce_u.h
index e3012e1..b3f21ba 100644
--- a/providers/hns/hns_roce_u.h
+++ b/providers/hns/hns_roce_u.h
@@ -277,6 +277,8 @@ struct hns_roce_context {
 	unsigned int			max_inline_data;
 
 	struct hns_roce_dca_ctx		dca_ctx;
+
+	bool reseted;
 };
 
 struct hns_roce_td {
@@ -309,6 +311,11 @@ struct hns_roce_cq {
 	unsigned long			flags;
 	unsigned int			cqe_size;
 	struct hns_roce_v2_cqe		*cqe;
+	struct list_head list_sq;
+	struct list_head list_rq;
+	struct list_head list_srq;
+	struct list_head list_xrc_srq;
+	struct hns_roce_v2_cqe *sw_cqe;
 };
 
 struct hns_roce_idx_que {
@@ -344,6 +351,7 @@ struct hns_roce_srq {
 	unsigned int			wqe_shift;
 	unsigned int			*db;
 	unsigned short			counter;
+	struct list_node xrc_srcq_node;
 };
 
 struct hns_roce_wq {
@@ -413,6 +421,10 @@ struct hns_roce_qp {
 	unsigned int			rb_sq_head; /* roll back sq head */
 	struct hns_roce_sge_info	sge_info;
 
+	struct list_node rcq_node;
+	struct list_node scq_node;
+	struct list_node srcq_node;
+
 	/* Just for UD. If not enabled, 'sl' in ibv_wc
 	 * will be filled with 'port_type' in cqe.
 	 */
diff --git a/providers/hns/hns_roce_u_hw_v2.c b/providers/hns/hns_roce_u_hw_v2.c
index 9238fe5..4e92397 100644
--- a/providers/hns/hns_roce_u_hw_v2.c
+++ b/providers/hns/hns_roce_u_hw_v2.c
@@ -843,6 +843,180 @@ static int hns_roce_poll_one(struct hns_roce_context *ctx,
 	return hns_roce_flush_cqe(*cur_qp, status);
 }
 
+static void hns_roce_fill_swc(struct hns_roce_cq *cq, struct ibv_wc *wc,
+			      uint64_t wr_id, uint32_t qp_num)
+{
+	if (!wc) {
+		cq->verbs_cq.cq_ex.status = IBV_WC_WR_FLUSH_ERR;
+		cq->verbs_cq.cq_ex.wr_id = wr_id;
+		hr_reg_write(cq->sw_cqe,  CQE_LCL_QPN, qp_num);
+		return;
+	}
+
+	wc->wr_id = wr_id;
+	wc->status = IBV_WC_WR_FLUSH_ERR;
+	wc->vendor_err = 0;
+	wc->qp_num = qp_num;
+}
+
+static int hns_roce_get_wq_swc(struct hns_roce_cq *cq, struct hns_roce_qp *qp,
+			       struct ibv_wc *wc, bool is_sq)
+{
+	struct hns_roce_wq *wq = is_sq ? &qp->sq : &qp->rq;
+	unsigned int left_wr;
+	uint64_t wr_id;
+
+	left_wr = wq->head - wq->tail;
+	if (left_wr == 0) {
+		if (is_sq)
+			list_del_init(&qp->scq_node);
+		else
+			list_del_init(&qp->rcq_node);
+
+		return -ENOENT;
+	}
+
+	wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
+	hns_roce_fill_swc(cq, wc, wr_id, qp->verbs_qp.qp.qp_num);
+	wq->tail++;
+	return V2_CQ_OK;
+}
+
+static int hns_roce_gen_sq_swc(struct hns_roce_cq *cq, struct ibv_wc *wc)
+{
+	struct hns_roce_qp *next, *qp = NULL;
+
+	list_for_each_safe(&cq->list_sq, qp, next, scq_node) {
+		if (hns_roce_get_wq_swc(cq, qp, wc, true) == -ENOENT)
+			continue;
+
+		return V2_CQ_OK;
+	}
+
+	return !wc ? -ENOENT : V2_CQ_EMPTY;
+}
+
+static int hns_roce_gen_rq_swc(struct hns_roce_cq *cq, struct ibv_wc *wc)
+{
+	struct hns_roce_qp *next, *qp = NULL;
+
+	list_for_each_safe(&cq->list_rq, qp, next, rcq_node) {
+		if (hns_roce_get_wq_swc(cq, qp, wc, false) == -ENOENT)
+			continue;
+
+		return V2_CQ_OK;
+	}
+
+	return !wc ? -ENOENT : V2_CQ_EMPTY;
+}
+
+static int hns_roce_get_srq_swc(struct hns_roce_cq *cq, struct hns_roce_qp *qp,
+				struct hns_roce_srq *srq, struct ibv_wc *wc)
+{
+	unsigned int left_wr;
+	uint64_t wr_id;
+
+	hns_roce_spin_lock(&srq->hr_lock);
+	left_wr = srq->idx_que.head - srq->idx_que.tail;
+	if (left_wr == 0) {
+		if (qp)
+			list_del_init(&qp->srcq_node);
+		else
+			list_del_init(&srq->xrc_srcq_node);
+
+		hns_roce_spin_unlock(&srq->hr_lock);
+		return -ENOENT;
+	}
+
+	wr_id = srq->wrid[srq->idx_que.tail & (srq->wqe_cnt - 1)];
+	hns_roce_fill_swc(cq, wc, wr_id, srq->srqn);
+	srq->idx_que.tail++;
+	hns_roce_spin_unlock(&srq->hr_lock);
+
+	return V2_CQ_OK;
+}
+
+static int hns_roce_gen_common_srq_swc(struct hns_roce_cq *cq,
+				       struct ibv_wc *wc)
+{
+	struct hns_roce_qp *next, *qp = NULL;
+	struct hns_roce_srq *srq;
+
+	list_for_each_safe(&cq->list_srq, qp, next, srcq_node) {
+		srq = to_hr_srq(qp->verbs_qp.qp.srq);
+		if (hns_roce_get_srq_swc(cq, qp, srq, wc) == -ENOENT)
+			continue;
+
+		return V2_CQ_OK;
+	}
+
+	return !wc ? -ENOENT : V2_CQ_EMPTY;
+}
+
+static int hns_roce_gen_xrc_srq_swc(struct hns_roce_cq *cq, struct ibv_wc *wc)
+{
+	struct hns_roce_srq *next, *srq = NULL;
+
+	list_for_each_safe(&cq->list_xrc_srq, srq, next, xrc_srcq_node) {
+		if (hns_roce_get_srq_swc(cq, NULL, srq, wc) == -ENOENT)
+			continue;
+
+		return V2_CQ_OK;
+	}
+
+	return !wc ? -ENOENT : V2_CQ_EMPTY;
+}
+
+static int hns_roce_gen_srq_swc(struct hns_roce_cq *cq, struct ibv_wc *wc)
+{
+	int err;
+
+	err = hns_roce_gen_common_srq_swc(cq, wc);
+	if (err == V2_CQ_OK)
+		return err;
+
+	return hns_roce_gen_xrc_srq_swc(cq, wc);
+}
+
+static int hns_roce_poll_one_swc(struct hns_roce_cq *cq, struct ibv_wc *wc)
+{
+	int err;
+
+	err = hns_roce_gen_sq_swc(cq, wc);
+	if (err == V2_CQ_OK)
+		return err;
+
+	err = hns_roce_gen_rq_swc(cq, wc);
+	if (err == V2_CQ_OK)
+		return err;
+
+	return hns_roce_gen_srq_swc(cq, wc);
+}
+
+static int hns_roce_poll_swc(struct hns_roce_cq *cq, int ne, struct ibv_wc *wc)
+{
+	int npolled;
+	int err;
+
+	for (npolled = 0; npolled < ne; npolled++) {
+		err = hns_roce_poll_one_swc(cq, wc + npolled);
+		if (err == V2_CQ_EMPTY)
+			break;
+	}
+
+	return npolled;
+}
+
+static bool hns_roce_reseted(struct hns_roce_context *ctx)
+{
+	struct hns_roce_v2_reset_state *state = ctx->reset_state;
+
+	if (state && state->is_reset)
+		ctx->reseted = true;
+
+	return ctx->reseted;
+}
+
 static int hns_roce_u_v2_poll_cq(struct ibv_cq *ibvcq, int ne,
 				 struct ibv_wc *wc)
 {
@@ -854,6 +1028,12 @@ static int hns_roce_u_v2_poll_cq(struct ibv_cq *ibvcq, int ne,
 
 	hns_roce_spin_lock(&cq->hr_lock);
 
+	if (unlikely(hns_roce_reseted(ctx))) {
+		npolled = hns_roce_poll_swc(cq, ne, wc);
+		hns_roce_spin_unlock(&cq->hr_lock);
+		return npolled;
+	}
+
 	for (npolled = 0; npolled < ne; ++npolled) {
 		err = hns_roce_poll_one(ctx, &qp, cq, wc + npolled);
 		if (qp && check_dca_detach_enable(qp))
@@ -1773,11 +1953,8 @@ static int hns_roce_u_v2_modify_qp(struct ibv_qp *qp, struct ibv_qp_attr *attr,
 	return ret;
 }
 
-static void hns_roce_lock_cqs(struct ibv_qp *qp)
+void hns_roce_lock_cqs(struct hns_roce_cq *send_cq, struct hns_roce_cq *recv_cq)
 {
-	struct hns_roce_cq *send_cq = to_hr_cq(qp->send_cq);
-	struct hns_roce_cq *recv_cq = to_hr_cq(qp->recv_cq);
-
 	if (send_cq && recv_cq) {
 		if (send_cq == recv_cq) {
 			hns_roce_spin_lock(&send_cq->hr_lock);
@@ -1795,11 +1972,8 @@ static void hns_roce_lock_cqs(struct ibv_qp *qp)
 	}
 }
 
-static void hns_roce_unlock_cqs(struct ibv_qp *qp)
+void hns_roce_unlock_cqs(struct hns_roce_cq *send_cq, struct hns_roce_cq *recv_cq)
 {
-	struct hns_roce_cq *send_cq = to_hr_cq(qp->send_cq);
-	struct hns_roce_cq *recv_cq = to_hr_cq(qp->recv_cq);
-
 	if (send_cq && recv_cq) {
 		if (send_cq == recv_cq) {
 			hns_roce_spin_unlock(&send_cq->hr_lock);
@@ -1832,17 +2006,22 @@ static int hns_roce_u_v2_destroy_qp(struct ibv_qp *ibqp)
 
 	hns_roce_v2_clear_qp(ctx, qp);
 
-	hns_roce_lock_cqs(ibqp);
+	hns_roce_lock_cqs(to_hr_cq(ibqp->send_cq), to_hr_cq(ibqp->recv_cq));
 
-	if (ibqp->recv_cq)
+	if (ibqp->recv_cq) {
 		__hns_roce_v2_cq_clean(to_hr_cq(ibqp->recv_cq), ibqp->qp_num,
 				       ibqp->srq ? to_hr_srq(ibqp->srq) : NULL);
+		list_del(&qp->srcq_node);
+		list_del(&qp->rcq_node);
+	}
 
-	if (ibqp->send_cq && ibqp->send_cq != ibqp->recv_cq)
+	if (ibqp->send_cq && ibqp->send_cq != ibqp->recv_cq) {
 		__hns_roce_v2_cq_clean(to_hr_cq(ibqp->send_cq), ibqp->qp_num,
 				       NULL);
+		list_del(&qp->scq_node);
+	}
 
-	hns_roce_unlock_cqs(ibqp);
+	hns_roce_unlock_cqs(to_hr_cq(ibqp->send_cq), to_hr_cq(ibqp->recv_cq));
 
 	hns_roce_free_qp_buf(qp, ctx);
 
@@ -1988,10 +2167,16 @@ static int wc_start_poll_cq(struct ibv_cq_ex *current,
 
 	hns_roce_spin_lock(&cq->hr_lock);
 
+	if (unlikely(hns_roce_reseted(ctx))) {
+		err = hns_roce_poll_one_swc(cq, NULL);
+		goto start_poll_done;
+	}
+
 	err = hns_roce_poll_one(ctx, &qp, cq, NULL);
 	if (qp && check_dca_detach_enable(qp))
 		dca_detach_qp_buf(ctx, qp);
 
+start_poll_done:
 	if (err != V2_CQ_OK)
 		hns_roce_spin_unlock(&cq->hr_lock);
 
@@ -2005,6 +2190,9 @@ static int wc_next_poll_cq(struct ibv_cq_ex *current)
 	struct hns_roce_qp *qp = NULL;
 	int err;
 
+	if (unlikely(hns_roce_reseted(ctx)))
+		return hns_roce_poll_one_swc(cq, NULL);
+
 	err = hns_roce_poll_one(ctx, &qp, cq, NULL);
 	if (qp && check_dca_detach_enable(qp))
 		dca_detach_qp_buf(ctx, qp);
@@ -2024,11 +2212,15 @@ static void wc_end_poll_cq(struct ibv_cq_ex *current)
 	struct hns_roce_cq *cq = to_hr_cq(ibv_cq_ex_to_cq(current));
 	struct hns_roce_context *ctx = to_hr_ctx(current->context);
 
+	if (unlikely(hns_roce_reseted(ctx)))
+		goto end_poll_done;
+
 	if (cq->flags & HNS_ROCE_CQ_FLAG_RECORD_DB)
 		*cq->db = cq->cons_index & RECORD_DB_CI_MASK;
 	else
 		update_cq_db(ctx, cq);
 
+end_poll_done:
 	hns_roce_spin_unlock(&cq->hr_lock);
 }
 
diff --git a/providers/hns/hns_roce_u_hw_v2.h b/providers/hns/hns_roce_u_hw_v2.h
index d628d76..50a920f 100644
--- a/providers/hns/hns_roce_u_hw_v2.h
+++ b/providers/hns/hns_roce_u_hw_v2.h
@@ -346,5 +346,7 @@ void hns_roce_v2_clear_qp(struct hns_roce_context *ctx, struct hns_roce_qp *qp);
 void hns_roce_attach_cq_ex_ops(struct ibv_cq_ex *cq_ex, uint64_t wc_flags);
 int hns_roce_attach_qp_ex_ops(struct ibv_qp_init_attr_ex *attr,
 			      struct hns_roce_qp *qp);
+void hns_roce_lock_cqs(struct hns_roce_cq *send_cq, struct hns_roce_cq *recv_cq);
+void hns_roce_unlock_cqs(struct hns_roce_cq *send_cq, struct hns_roce_cq *recv_cq);
 
 #endif /* _HNS_ROCE_U_HW_V2_H */
diff --git a/providers/hns/hns_roce_u_verbs.c b/providers/hns/hns_roce_u_verbs.c
index 4b641ea..8fb415b 100644
--- a/providers/hns/hns_roce_u_verbs.c
+++ b/providers/hns/hns_roce_u_verbs.c
@@ -519,6 +519,32 @@ static int exec_cq_create_cmd(struct ibv_context *context,
 	return 0;
 }
 
+static int hns_roce_init_cq_swc(struct hns_roce_cq *cq,
+				struct ibv_cq_init_attr_ex *attr)
+{
+	list_head_init(&cq->list_sq);
+	list_head_init(&cq->list_rq);
+	list_head_init(&cq->list_srq);
+	list_head_init(&cq->list_xrc_srq);
+
+	if (!(attr->wc_flags & CREATE_CQ_SUPPORTED_WC_FLAGS))
+		return 0;
+
+	cq->sw_cqe = calloc(1, sizeof(struct hns_roce_v2_cqe));
+	if (!cq->sw_cqe)
+		return -ENOMEM;
+
+	return 0;
+}
+
+static void hns_roce_uninit_cq_swc(struct hns_roce_cq *cq)
+{
+	if (cq->sw_cqe) {
+		free(cq->sw_cqe);
+		cq->sw_cqe = NULL;
+	}
+}
+
 static struct ibv_cq_ex *create_cq(struct ibv_context *context,
 			 struct ibv_cq_init_attr_ex *attr)
 {
@@ -564,6 +590,10 @@ static struct ibv_cq_ex *create_cq(struct ibv_context *context,
 
 	*cq->db = 0;
 
+	ret = hns_roce_init_cq_swc(cq, attr);
+	if (ret)
+		goto err_swc;
+
 	ret = exec_cq_create_cmd(context, cq, attr);
 	if (ret)
 		goto err_cmd;
@@ -573,6 +603,8 @@ static struct ibv_cq_ex *create_cq(struct ibv_context *context,
 	return &cq->verbs_cq.cq_ex;
 
 err_cmd:
+	hns_roce_uninit_cq_swc(cq);
+err_swc:
 	hns_roce_free_db(hr_ctx, cq->db, HNS_ROCE_CQ_TYPE_DB);
 err_db:
 	hns_roce_free_buf(&cq->buf);
@@ -632,6 +664,8 @@ int hns_roce_u_destroy_cq(struct ibv_cq *cq)
 	if (ret)
 		return ret;
 
+	hns_roce_uninit_cq_swc(to_hr_cq(cq));
+
 	hns_roce_free_db(to_hr_ctx(cq->context), to_hr_cq(cq)->db,
 				   HNS_ROCE_CQ_TYPE_DB);
 	hns_roce_free_buf(&to_hr_cq(cq)->buf);
@@ -839,6 +873,22 @@ static int exec_srq_create_cmd(struct ibv_context *context,
 	return 0;
 }
 
+static void init_srq_cq_list(struct hns_roce_srq *srq,
+			     struct ibv_srq_init_attr_ex *init_attr)
+{
+	struct hns_roce_cq *srq_cq;
+
+	list_node_init(&srq->xrc_srcq_node);
+
+	if (!init_attr->cq)
+		return;
+
+	srq_cq = to_hr_cq(init_attr->cq);
+	hns_roce_spin_lock(&srq_cq->hr_lock);
+	list_add_tail(&srq_cq->list_xrc_srq, &srq->xrc_srcq_node);
+	hns_roce_spin_unlock(&srq_cq->hr_lock);
+}
+
 static struct ibv_srq *create_srq(struct ibv_context *context,
 				  struct ibv_srq_init_attr_ex *init_attr)
 {
@@ -886,6 +936,8 @@ static struct ibv_srq *create_srq(struct ibv_context *context,
 	init_attr->attr.max_sge =
 		min(init_attr->attr.max_sge - srq->rsv_sge, hr_ctx->max_srq_sge);
 
+	init_srq_cq_list(srq, init_attr);
+
 	return &srq->verbs_srq.srq;
 
 err_destroy_srq:
@@ -958,12 +1010,26 @@ int hns_roce_u_query_srq(struct ibv_srq *srq, struct ibv_srq_attr *srq_attr)
 	return ret;
 }
 
+static void del_srq_from_cq_list(struct hns_roce_srq *srq)
+{
+	struct hns_roce_cq *srq_cq = to_hr_cq(srq->verbs_srq.cq);
+
+	if (!srq_cq)
+		return;
+
+	hns_roce_spin_lock(&srq_cq->hr_lock);
+	list_del(&srq->xrc_srcq_node);
+	hns_roce_spin_unlock(&srq_cq->hr_lock);
+}
+
 int hns_roce_u_destroy_srq(struct ibv_srq *ibv_srq)
 {
 	struct hns_roce_context *ctx = to_hr_ctx(ibv_srq->context);
 	struct hns_roce_srq *srq = to_hr_srq(ibv_srq);
 	int ret;
 
+	del_srq_from_cq_list(srq);
+
 	ret = ibv_cmd_destroy_srq(ibv_srq);
 	if (ret)
 		return ret;
@@ -1648,6 +1714,30 @@ static int mmap_dwqe(struct ibv_context *ibv_ctx, struct hns_roce_qp *qp,
 	return 0;
 }
 
+static void add_qp_to_cq_list(struct ibv_qp_init_attr_ex *attr,
+			      struct hns_roce_qp *qp)
+{
+	struct hns_roce_cq *send_cq, *recv_cq;
+
+	send_cq = attr->send_cq ? to_hr_cq(attr->send_cq) : NULL;
+	recv_cq = attr->recv_cq ? to_hr_cq(attr->recv_cq) : NULL;
+
+	list_node_init(&qp->scq_node);
+	list_node_init(&qp->rcq_node);
+	list_node_init(&qp->srcq_node);
+
+	hns_roce_lock_cqs(send_cq, recv_cq);
+	if (send_cq)
+		list_add_tail(&send_cq->list_sq, &qp->scq_node);
+	if (recv_cq) {
+		if (attr->srq)
+			list_add_tail(&recv_cq->list_srq, &qp->srcq_node);
+		else
+			list_add_tail(&recv_cq->list_rq, &qp->rcq_node);
+	}
+	hns_roce_unlock_cqs(send_cq, recv_cq);
+}
+
 static struct ibv_qp *create_qp(struct ibv_context *ibv_ctx,
 				struct ibv_qp_init_attr_ex *attr,
 				struct hnsdv_qp_init_attr *hns_attr)
@@ -1697,6 +1787,7 @@ static struct ibv_qp *create_qp(struct ibv_context *ibv_ctx,
 	}
 
 	qp_setup_config(attr, qp, context);
+	add_qp_to_cq_list(attr, qp);
 
 	if (hns_attr && hns_attr->create_flags & HNSDV_QP_CREATE_ENABLE_UD_SL)
 		qp->enable_ud_sl = true;
-- 
2.25.1