RDMA/i40iw: Correct QP size calculation

Include inline data size as part of SQ size calculation.
RQ size calculation uses only number of SGEs and does not
support 96 byte WQE size.

Signed-off-by: Mustafa Ismail <mustafa.ismail@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

4 files changed, 58 insertions, 66 deletions

diff --git a/drivers/infiniband/hw/i40iw/i40iw_d.h b/drivers/infiniband/hw/i40iw/i40iw_d.h
index aab88d65f805..e8951a71cc13 100644
--- a/drivers/infiniband/hw/i40iw/i40iw_d.h
+++ b/drivers/infiniband/hw/i40iw/i40iw_d.h
@@ -1290,7 +1290,7 @@
 
 /* wqe size considering 32 bytes per wqe*/
 #define I40IWQP_SW_MIN_WQSIZE 4		/* 128 bytes */
-#define I40IWQP_SW_MAX_WQSIZE 16384	/* 524288 bytes */
+#define I40IWQP_SW_MAX_WQSIZE 2048	/* 2048 bytes */
 
 #define I40IWQP_OP_RDMA_WRITE 0
 #define I40IWQP_OP_RDMA_READ 1
diff --git a/drivers/infiniband/hw/i40iw/i40iw_uk.c b/drivers/infiniband/hw/i40iw/i40iw_uk.c
index f78c3dc8bdb2..9e3a700d5a2d 100644
--- a/drivers/infiniband/hw/i40iw/i40iw_uk.c
+++ b/drivers/infiniband/hw/i40iw/i40iw_uk.c
@@ -130,7 +130,10 @@ static void i40iw_qp_ring_push_db(struct i40iw_qp_uk *qp, u32 wqe_idx)
  */
 u64 *i40iw_qp_get_next_send_wqe(struct i40iw_qp_uk *qp,
 				u32 *wqe_idx,
-				u8 wqe_size)
+				u8 wqe_size,
+				u32 total_size,
+				u64 wr_id
+				)
 {
 	u64 *wqe = NULL;
 	u64 wqe_ptr;
@@ -171,6 +174,10 @@ u64 *i40iw_qp_get_next_send_wqe(struct i40iw_qp_uk *qp,
 	wqe_0 = qp->sq_base[peek_head].elem;
 	if (peek_head & 0x3)
 		wqe_0[3] = LS_64(!qp->swqe_polarity, I40IWQPSQ_VALID);
+
+	qp->sq_wrtrk_array[*wqe_idx].wrid = wr_id;
+	qp->sq_wrtrk_array[*wqe_idx].wr_len = total_size;
+	qp->sq_wrtrk_array[*wqe_idx].wqe_size = wqe_size;
 	return wqe;
 }
 
@@ -249,12 +256,9 @@ static enum i40iw_status_code i40iw_rdma_write(struct i40iw_qp_uk *qp,
 	if (ret_code)
 		return ret_code;
 
-	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, wqe_size);
+	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, wqe_size, total_size, info->wr_id);
 	if (!wqe)
 		return I40IW_ERR_QP_TOOMANY_WRS_POSTED;
-
-	qp->sq_wrtrk_array[wqe_idx].wrid = info->wr_id;
-	qp->sq_wrtrk_array[wqe_idx].wr_len = total_size;
 	set_64bit_val(wqe, 16,
 		      LS_64(op_info->rem_addr.tag_off, I40IWQPSQ_FRAG_TO));
 	if (!op_info->rem_addr.stag)
@@ -309,12 +313,9 @@ static enum i40iw_status_code i40iw_rdma_read(struct i40iw_qp_uk *qp,
 	ret_code = i40iw_fragcnt_to_wqesize_sq(1, &wqe_size);
 	if (ret_code)
 		return ret_code;
-	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, wqe_size);
+	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, wqe_size, op_info->lo_addr.len, info->wr_id);
 	if (!wqe)
 		return I40IW_ERR_QP_TOOMANY_WRS_POSTED;
-
-	qp->sq_wrtrk_array[wqe_idx].wrid = info->wr_id;
-	qp->sq_wrtrk_array[wqe_idx].wr_len = op_info->lo_addr.len;
 	local_fence |= info->local_fence;
 
 	set_64bit_val(wqe, 16, LS_64(op_info->rem_addr.tag_off, I40IWQPSQ_FRAG_TO));
@@ -366,13 +367,11 @@ static enum i40iw_status_code i40iw_send(struct i40iw_qp_uk *qp,
 	if (ret_code)
 		return ret_code;
 
-	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, wqe_size);
+	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, wqe_size, total_size, info->wr_id);
 	if (!wqe)
 		return I40IW_ERR_QP_TOOMANY_WRS_POSTED;
 
 	read_fence |= info->read_fence;
-	qp->sq_wrtrk_array[wqe_idx].wrid = info->wr_id;
-	qp->sq_wrtrk_array[wqe_idx].wr_len = total_size;
 	set_64bit_val(wqe, 16, 0);
 	header = LS_64(stag_to_inv, I40IWQPSQ_REMSTAG) |
 		 LS_64(info->op_type, I40IWQPSQ_OPCODE) |
@@ -427,13 +426,11 @@ static enum i40iw_status_code i40iw_inline_rdma_write(struct i40iw_qp_uk *qp,
 	if (ret_code)
 		return ret_code;
 
-	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, wqe_size);
+	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, wqe_size, op_info->len, info->wr_id);
 	if (!wqe)
 		return I40IW_ERR_QP_TOOMANY_WRS_POSTED;
 
 	read_fence |= info->read_fence;
-	qp->sq_wrtrk_array[wqe_idx].wrid = info->wr_id;
-	qp->sq_wrtrk_array[wqe_idx].wr_len = op_info->len;
 	set_64bit_val(wqe, 16,
 		      LS_64(op_info->rem_addr.tag_off, I40IWQPSQ_FRAG_TO));
 
@@ -507,14 +504,11 @@ static enum i40iw_status_code i40iw_inline_send(struct i40iw_qp_uk *qp,
 	if (ret_code)
 		return ret_code;
 
-	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, wqe_size);
+	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, wqe_size, op_info->len, info->wr_id);
 	if (!wqe)
 		return I40IW_ERR_QP_TOOMANY_WRS_POSTED;
 
 	read_fence |= info->read_fence;
-
-	qp->sq_wrtrk_array[wqe_idx].wrid = info->wr_id;
-	qp->sq_wrtrk_array[wqe_idx].wr_len = op_info->len;
 	header = LS_64(stag_to_inv, I40IWQPSQ_REMSTAG) |
 	    LS_64(info->op_type, I40IWQPSQ_OPCODE) |
 	    LS_64(op_info->len, I40IWQPSQ_INLINEDATALEN) |
@@ -574,12 +568,9 @@ static enum i40iw_status_code i40iw_stag_local_invalidate(struct i40iw_qp_uk *qp
 	op_info = &info->op.inv_local_stag;
 	local_fence = info->local_fence;
 
-	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, I40IW_QP_WQE_MIN_SIZE);
+	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, I40IW_QP_WQE_MIN_SIZE, 0, info->wr_id);
 	if (!wqe)
 		return I40IW_ERR_QP_TOOMANY_WRS_POSTED;
-
-	qp->sq_wrtrk_array[wqe_idx].wrid = info->wr_id;
-	qp->sq_wrtrk_array[wqe_idx].wr_len = 0;
 	set_64bit_val(wqe, 0, 0);
 	set_64bit_val(wqe, 8,
 		      LS_64(op_info->target_stag, I40IWQPSQ_LOCSTAG));
@@ -619,12 +610,9 @@ static enum i40iw_status_code i40iw_mw_bind(struct i40iw_qp_uk *qp,
 	op_info = &info->op.bind_window;
 
 	local_fence |= info->local_fence;
-	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, I40IW_QP_WQE_MIN_SIZE);
+	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, I40IW_QP_WQE_MIN_SIZE, 0, info->wr_id);
 	if (!wqe)
 		return I40IW_ERR_QP_TOOMANY_WRS_POSTED;
-
-	qp->sq_wrtrk_array[wqe_idx].wrid = info->wr_id;
-	qp->sq_wrtrk_array[wqe_idx].wr_len = 0;
 	set_64bit_val(wqe, 0, (uintptr_t)op_info->va);
 	set_64bit_val(wqe, 8,
 		      LS_64(op_info->mr_stag, I40IWQPSQ_PARENTMRSTAG) |
@@ -760,7 +748,7 @@ static enum i40iw_status_code i40iw_cq_poll_completion(struct i40iw_cq_uk *cq,
 	enum i40iw_status_code ret_code2 = 0;
 	bool move_cq_head = true;
 	u8 polarity;
-	u8 addl_frag_cnt, addl_wqes = 0;
+	u8 addl_wqes = 0;
 
 	if (cq->avoid_mem_cflct)
 		cqe = (u64 *)I40IW_GET_CURRENT_EXTENDED_CQ_ELEMENT(cq);
@@ -827,11 +815,8 @@ static enum i40iw_status_code i40iw_cq_poll_completion(struct i40iw_cq_uk *cq,
 			info->op_type = (u8)RS_64(qword3, I40IWCQ_OP);
 			sw_wqe = qp->sq_base[wqe_idx].elem;
 			get_64bit_val(sw_wqe, 24, &wqe_qword);
-			addl_frag_cnt =
-			    (u8)RS_64(wqe_qword, I40IWQPSQ_ADDFRAGCNT);
-			i40iw_fragcnt_to_wqesize_sq(addl_frag_cnt + 1, &addl_wqes);
 
-			addl_wqes = (addl_wqes / I40IW_QP_WQE_MIN_SIZE);
+			addl_wqes = qp->sq_wrtrk_array[wqe_idx].wqe_size / I40IW_QP_WQE_MIN_SIZE;
 			I40IW_RING_SET_TAIL(qp->sq_ring, (wqe_idx + addl_wqes));
 		} else {
 			do {
@@ -843,9 +828,7 @@ static enum i40iw_status_code i40iw_cq_poll_completion(struct i40iw_cq_uk *cq,
 				get_64bit_val(sw_wqe, 24, &wqe_qword);
 				op_type = (u8)RS_64(wqe_qword, I40IWQPSQ_OPCODE);
 				info->op_type = op_type;
-				addl_frag_cnt = (u8)RS_64(wqe_qword, I40IWQPSQ_ADDFRAGCNT);
-				i40iw_fragcnt_to_wqesize_sq(addl_frag_cnt + 1, &addl_wqes);
-				addl_wqes = (addl_wqes / I40IW_QP_WQE_MIN_SIZE);
+				addl_wqes = qp->sq_wrtrk_array[tail].wqe_size / I40IW_QP_WQE_MIN_SIZE;
 				I40IW_RING_SET_TAIL(qp->sq_ring, (tail + addl_wqes));
 				if (op_type != I40IWQP_OP_NOP) {
 					info->wr_id = qp->sq_wrtrk_array[tail].wrid;
@@ -893,19 +876,21 @@ static enum i40iw_status_code i40iw_cq_poll_completion(struct i40iw_cq_uk *cq,
  * i40iw_get_wqe_shift - get shift count for maximum wqe size
  * @wqdepth: depth of wq required.
  * @sge: Maximum Scatter Gather Elements wqe
+ * @inline_data: Maximum inline data size
  * @shift: Returns the shift needed based on sge
  *
- * Shift can be used to left shift the wqe size based on sge.
- * If sge, == 1, shift =0 (wqe_size of 32 bytes), for sge=2 and 3, shift =1
- * (64 bytes wqes) and 2 otherwise (128 bytes wqe).
+ * Shift can be used to left shift the wqe size based on number of SGEs and inlind data size.
+ * For 1 SGE or inline data <= 16, shift = 0 (wqe size of 32 bytes).
+ * For 2 or 3 SGEs or inline data <= 48, shift = 1 (wqe size of 64 bytes).
+ * Shift of 2 otherwise (wqe size of 128 bytes).
  */
-enum i40iw_status_code i40iw_get_wqe_shift(u32 wqdepth, u8 sge, u8 *shift)
+enum i40iw_status_code i40iw_get_wqe_shift(u32 wqdepth, u32 sge, u32 inline_data, u8 *shift)
 {
 	u32 size;
 
 	*shift = 0;
-	if (sge > 1)
-		*shift = (sge < 4) ? 1 : 2;
+	if (sge > 1 || inline_data > 16)
+		*shift = (sge < 4 && inline_data <= 48) ? 1 : 2;
 
 	/* check if wqdepth is multiple of 2 or not */
 
@@ -968,11 +953,11 @@ enum i40iw_status_code i40iw_qp_uk_init(struct i40iw_qp_uk *qp,
 
 	if (info->max_rq_frag_cnt > I40IW_MAX_WQ_FRAGMENT_COUNT)
 		return I40IW_ERR_INVALID_FRAG_COUNT;
-	ret_code = i40iw_get_wqe_shift(info->sq_size, info->max_sq_frag_cnt, &sqshift);
+	ret_code = i40iw_get_wqe_shift(info->sq_size, info->max_sq_frag_cnt, info->max_inline_data, &sqshift);
 	if (ret_code)
 		return ret_code;
 
-	ret_code = i40iw_get_wqe_shift(info->rq_size, info->max_rq_frag_cnt, &rqshift);
+	ret_code = i40iw_get_wqe_shift(info->rq_size, info->max_rq_frag_cnt, 0, &rqshift);
 	if (ret_code)
 		return ret_code;
 
@@ -1097,12 +1082,9 @@ enum i40iw_status_code i40iw_nop(struct i40iw_qp_uk *qp,
 	u64 header, *wqe;
 	u32 wqe_idx;
 
-	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, I40IW_QP_WQE_MIN_SIZE);
+	wqe = i40iw_qp_get_next_send_wqe(qp, &wqe_idx, I40IW_QP_WQE_MIN_SIZE, 0, wr_id);
 	if (!wqe)
 		return I40IW_ERR_QP_TOOMANY_WRS_POSTED;
-
-	qp->sq_wrtrk_array[wqe_idx].wrid = wr_id;
-	qp->sq_wrtrk_array[wqe_idx].wr_len = 0;
 	set_64bit_val(wqe, 0, 0);
 	set_64bit_val(wqe, 8, 0);
 	set_64bit_val(wqe, 16, 0);
@@ -1125,7 +1107,7 @@ enum i40iw_status_code i40iw_nop(struct i40iw_qp_uk *qp,
  * @frag_cnt: number of fragments
  * @wqe_size: size of sq wqe returned
  */
-enum i40iw_status_code i40iw_fragcnt_to_wqesize_sq(u8 frag_cnt, u8 *wqe_size)
+enum i40iw_status_code i40iw_fragcnt_to_wqesize_sq(u32 frag_cnt, u8 *wqe_size)
 {
 	switch (frag_cnt) {
 	case 0:
@@ -1156,7 +1138,7 @@ enum i40iw_status_code i40iw_fragcnt_to_wqesize_sq(u8 frag_cnt, u8 *wqe_size)
  * @frag_cnt: number of fragments
  * @wqe_size: size of rq wqe returned
  */
-enum i40iw_status_code i40iw_fragcnt_to_wqesize_rq(u8 frag_cnt, u8 *wqe_size)
+enum i40iw_status_code i40iw_fragcnt_to_wqesize_rq(u32 frag_cnt, u8 *wqe_size)
 {
 	switch (frag_cnt) {
 	case 0:
diff --git a/drivers/infiniband/hw/i40iw/i40iw_user.h b/drivers/infiniband/hw/i40iw/i40iw_user.h
index eac95240fbdc..4627646fe8cd 100644
--- a/drivers/infiniband/hw/i40iw/i40iw_user.h
+++ b/drivers/infiniband/hw/i40iw/i40iw_user.h
@@ -61,7 +61,7 @@ enum i40iw_device_capabilities_const {
 	I40IW_MAX_CQ_SIZE =			1048575,
 	I40IW_MAX_AEQ_ALLOCATE_COUNT =		255,
 	I40IW_DB_ID_ZERO =			0,
-	I40IW_MAX_WQ_FRAGMENT_COUNT =		6,
+	I40IW_MAX_WQ_FRAGMENT_COUNT =		3,
 	I40IW_MAX_SGE_RD =			1,
 	I40IW_MAX_OUTBOUND_MESSAGE_SIZE =	2147483647,
 	I40IW_MAX_INBOUND_MESSAGE_SIZE =	2147483647,
@@ -70,8 +70,8 @@ enum i40iw_device_capabilities_const {
 	I40IW_MAX_VF_FPM_ID =			47,
 	I40IW_MAX_VF_PER_PF =			127,
 	I40IW_MAX_SQ_PAYLOAD_SIZE =		2145386496,
-	I40IW_MAX_INLINE_DATA_SIZE =		112,
-	I40IW_MAX_PUSHMODE_INLINE_DATA_SIZE =	112,
+	I40IW_MAX_INLINE_DATA_SIZE =		48,
+	I40IW_MAX_PUSHMODE_INLINE_DATA_SIZE =	48,
 	I40IW_MAX_IRD_SIZE =			32,
 	I40IW_QPCTX_ENCD_MAXIRD =		3,
 	I40IW_MAX_WQ_ENTRIES =			2048,
@@ -200,7 +200,7 @@ enum i40iw_completion_notify {
 
 struct i40iw_post_send {
 	i40iw_sgl sg_list;
-	u8 num_sges;
+	u32 num_sges;
 };
 
 struct i40iw_post_inline_send {
@@ -222,7 +222,7 @@ struct i40iw_post_inline_send_w_inv {
 
 struct i40iw_rdma_write {
 	i40iw_sgl lo_sg_list;
-	u8 num_lo_sges;
+	u32 num_lo_sges;
 	struct i40iw_sge rem_addr;
 };
 
@@ -347,7 +347,9 @@ struct i40iw_dev_uk {
 
 struct i40iw_sq_uk_wr_trk_info {
 	u64 wrid;
-	u64 wr_len;
+	u32 wr_len;
+	u8 wqe_size;
+	u8 reserved[3];
 };
 
 struct i40iw_qp_quanta {
@@ -369,6 +371,8 @@ struct i40iw_qp_uk {
 	u32 qp_id;
 	u32 sq_size;
 	u32 rq_size;
+	u32 max_sq_frag_cnt;
+	u32 max_rq_frag_cnt;
 	struct i40iw_qp_uk_ops ops;
 	bool use_srq;
 	u8 swqe_polarity;
@@ -376,8 +380,6 @@ struct i40iw_qp_uk {
 	u8 rwqe_polarity;
 	u8 rq_wqe_size;
 	u8 rq_wqe_size_multiplier;
-	u8 max_sq_frag_cnt;
-	u8 max_rq_frag_cnt;
 	bool deferred_flag;
 };
 
@@ -406,8 +408,9 @@ struct i40iw_qp_uk_init_info {
 	u32 qp_id;
 	u32 sq_size;
 	u32 rq_size;
-	u8 max_sq_frag_cnt;
-	u8 max_rq_frag_cnt;
+	u32 max_sq_frag_cnt;
+	u32 max_rq_frag_cnt;
+	u32 max_inline_data;
 
 };
 
@@ -424,7 +427,10 @@ void i40iw_device_init_uk(struct i40iw_dev_uk *dev);
 
 void i40iw_qp_post_wr(struct i40iw_qp_uk *qp);
 u64 *i40iw_qp_get_next_send_wqe(struct i40iw_qp_uk *qp, u32 *wqe_idx,
-				u8 wqe_size);
+				u8 wqe_size,
+				u32 total_size,
+				u64 wr_id
+				);
 u64 *i40iw_qp_get_next_recv_wqe(struct i40iw_qp_uk *qp, u32 *wqe_idx);
 u64 *i40iw_qp_get_next_srq_wqe(struct i40iw_srq_uk *srq, u32 *wqe_idx);
 
@@ -436,9 +442,9 @@ enum i40iw_status_code i40iw_qp_uk_init(struct i40iw_qp_uk *qp,
 void i40iw_clean_cq(void *queue, struct i40iw_cq_uk *cq);
 enum i40iw_status_code i40iw_nop(struct i40iw_qp_uk *qp, u64 wr_id,
 				 bool signaled, bool post_sq);
-enum i40iw_status_code i40iw_fragcnt_to_wqesize_sq(u8 frag_cnt, u8 *wqe_size);
-enum i40iw_status_code i40iw_fragcnt_to_wqesize_rq(u8 frag_cnt, u8 *wqe_size);
+enum i40iw_status_code i40iw_fragcnt_to_wqesize_sq(u32 frag_cnt, u8 *wqe_size);
+enum i40iw_status_code i40iw_fragcnt_to_wqesize_rq(u32 frag_cnt, u8 *wqe_size);
 enum i40iw_status_code i40iw_inline_data_size_to_wqesize(u32 data_size,
 							 u8 *wqe_size);
-enum i40iw_status_code i40iw_get_wqe_shift(u32 wqdepth, u8 sge, u8 *shift);
+enum i40iw_status_code i40iw_get_wqe_shift(u32 wqdepth, u32 sge, u32 inline_data, u8 *shift);
 #endif
diff --git a/drivers/infiniband/hw/i40iw/i40iw_verbs.c b/drivers/infiniband/hw/i40iw/i40iw_verbs.c
index d7c4dd15f1c0..329f59a9f18a 100644
--- a/drivers/infiniband/hw/i40iw/i40iw_verbs.c
+++ b/drivers/infiniband/hw/i40iw/i40iw_verbs.c
@@ -526,9 +526,9 @@ static int i40iw_setup_kmode_qp(struct i40iw_device *iwdev,
 	sq_size = i40iw_qp_roundup(ukinfo->sq_size + 1);
 	rq_size = i40iw_qp_roundup(ukinfo->rq_size + 1);
 
-	status = i40iw_get_wqe_shift(sq_size, ukinfo->max_sq_frag_cnt, &sqshift);
+	status = i40iw_get_wqe_shift(sq_size, ukinfo->max_sq_frag_cnt, ukinfo->max_inline_data, &sqshift);
 	if (!status)
-		status = i40iw_get_wqe_shift(rq_size, ukinfo->max_rq_frag_cnt, &rqshift);
+		status = i40iw_get_wqe_shift(rq_size, ukinfo->max_rq_frag_cnt, 0, &rqshift);
 
 	if (status)
 		return -ENOSYS;
@@ -609,6 +609,9 @@ static struct ib_qp *i40iw_create_qp(struct ib_pd *ibpd,
 	if (init_attr->cap.max_inline_data > I40IW_MAX_INLINE_DATA_SIZE)
 		init_attr->cap.max_inline_data = I40IW_MAX_INLINE_DATA_SIZE;
 
+	if (init_attr->cap.max_send_sge > I40IW_MAX_WQ_FRAGMENT_COUNT)
+		init_attr->cap.max_send_sge = I40IW_MAX_WQ_FRAGMENT_COUNT;
+
 	memset(&init_info, 0, sizeof(init_info));
 
 	sq_size = init_attr->cap.max_send_wr;
@@ -618,6 +621,7 @@ static struct ib_qp *i40iw_create_qp(struct ib_pd *ibpd,
 	init_info.qp_uk_init_info.rq_size = rq_size;
 	init_info.qp_uk_init_info.max_sq_frag_cnt = init_attr->cap.max_send_sge;
 	init_info.qp_uk_init_info.max_rq_frag_cnt = init_attr->cap.max_recv_sge;
+	init_info.qp_uk_init_info.max_inline_data = init_attr->cap.max_inline_data;
 
 	mem = kzalloc(sizeof(*iwqp), GFP_KERNEL);
 	if (!mem)